A Survey On Routing Protocols That Really Exploit Wireless Mesh Network Features

JOURNAL OF COMMUNICATIONS, VOL. 5, NO.
3, MARCH 2010
211
A Survey on Routing Protocols that really Exploit

Wireless Mesh Network Features
Jos Nez-Martnez, Josep Mangues-Bafalluy
Centre Tecnolgic de Telecomunicacions de Catalunya (CTTC)

Parc Mediterrani de la Tecnologia Av. Carl Friedrich Gauss, 7 08860 Castelldefels Barcelona Spain
Tel: +34 936452900, Fax: +34 936452901
Email: {jose.nunez, josep.mangues}@cttc.cat
Abstract The design of a routing protocol that really
exploits the specific features of a Wireless Mesh

Network (WMN) still remains a challenge. The
expected additional benefit, with respect to those that
do not or cannot exploit them, is an increase in overall
throughput supported by the network. The static and
non-power constrained nature of backbone nodes
allow offering some exploitable features towards this
goal, such as multi-radio and multi-channel support,
stability, and increased CPU and storage capabilities.
Such features have a strong impact on the design of
the routing scheme, which makes inefficient to port
existing solutions from wired and other wireless
networks. As a consequence, in recent years, many
alternative routing protocols have been proposed for
WMNs. This survey paper presents the current stateof-the-art of routing protocols specifically designed
for WMNs that try to maximize the throughput
transferred by the network. First, a classification of
routing protocols is provided. Second, the routing
architecture is decomposed into three major building
blocks. Open research issues related to each building
block are also discussed. And finally, the main
characteristics of the building blocks for each relevant
routing protocol in the taxonomy are summarized.
Index Termsrouting, wireless mesh network, multi-radio,
multi-channel, multi-rate, opportunistic routing, overview,
network coding
I. INTRODUCTION
WMNs [25] have various application scenarios, such
as building automation, wireless community networks, or
providing an easy and economical last-mile Internet
access. Instead of deploying costly wired infrastructure,
WMNs offer a low-cost multi-hop wireless backbone.
This has generated considerable interest from industry,
e.g. [27] or [28].
However, there are still several issues to be resolved.
One of the major problems to be tackled is the design of
an optimal routing protocol. In fact, although many
routing protocols from wired and other wireless networks
can be deployed in WMNs, they do not fully exploit their
singular characteristics.
On the one hand, wired-based routing protocols
assume each link in the network can only be in two states
Manuscript received May 25, 2009; revised Sept. 10, 2009; accepted
Nov. 17, 2009.
2010 ACADEMY PUBLISHER

doi:10.4304/jcm.5.3.211-231
in terms of connectivity: either the wireless link works

perfectly or it does not work at all. Additionally, they
usually rely on the hop-count metric, which is not
suitable for WMNs. Shortest paths in terms of number of
hops often lead to the selection of paths composed of bad
quality links. The reason for this is that selected links are
the longest ones in terms of distance, hence decreasing
the received signal-to-noise ratio (SNR) at the nodes and
generating more interference to neighboring nodes using
the same channel. Moreover, in protocols for wired
networks, the typical volume of routing overhead is not a
concern, since they rely on links in the order of Gigabits
per second.
However, wireless links cannot be compared with
wired links in terms of bandwidth, even under ideal
wireless channel conditions i.e., perfect line of sight and
no interference. Furthermore, wireless links are subject to
high variability. But despite all this, WMNs are expected
to allow running high throughput Internet applications.
Therefore, the design of high throughput routing
protocols able to appropriately deal with wireless links
becomes a necessity.
In wireless ad-hoc networks, the main design goal of
routing protocols is end-to-end connectivity maintenance.
Moreover, transmitting a packet is highly costly due to
node energy constraints. Therefore, at the routing level,
the research is mainly focused on designing low control
overhead routing techniques to deal with frequent route
breaks due to mobility or node failures.
By contrast, with emerging WMNs, the efforts towards
the design of routing protocols have been shifted from
merely maintaining connectivity to obtaining high
throughput end-to-end paths [40]. The WMN backbone is
composed of static and non-power constrained Wireless
Mesh Routers (WMRs). As a consequence, mobility and
energy limitations are not a concern anymore for WMRs
[24]. Then, besides maintaining, amongst others,
reliability and scalability as desirable properties, the main
new target for a routing protocol for WMNs is throughput
maximization, even at the expense of generating more
routing overhead, which deserves a secondary role [40].
This paper provides a comprehensive survey of
practical WMN routing protocols for throughput
maximization by considering the unique properties of
WMNs. As stated in [40], the main performance metric
for evaluating a routing protocol for WMNs is
212
throughput. Therefore, the main goal of a routing

protocol for WMNs should be to maximize the
throughput received by the destination of a single flow or
the overall throughput of the WMN. In this paper, we
focus on routing protocols for WMNs that have
throughput as their primary metric/target for measuring
their performance.
In addition, the routing protocols presented in this
paper are implementable in the sense that we restrict it to
those protocols that have been fully implemented in a real
testbed or in a network simulator.
We provide a classification of routing protocols based
on the characteristics exploited that are unique to WMNs.
Additionally, we also identify the three building blocks a
unicast routing protocol for WMNs may be decomposed
into. Open research issues that future routing protocols
may consider when designing each building block are
also discussed.
The classification proposed is orthogonal to state-ofthe-art classification approaches. Existing recent surveys
in the literature are not specifically focused on WMN
characteristics ([30], [36],) or at least, the WMNs
characteristics taken into account in this work. By
contrast, in this paper, the criteria for grouping routing
protocols are unique to WMNs. Moreover, this paper
presents what is, up to our knowledge, the first study of
research work specifically developed for WMNs for each
of the building blocks a routing protocol is composed of.
The rest of this paper is organized as follows. Section
II outlines the main limitations and advantages of WMNs.
Section III summarizes the key decisions in the design of
a routing protocol. Section IV presents a novel way to
classify routing protocols based on the WMN properties
exploited to maximize throughput. Section V decomposes
the routing architecture into building blocks and identifies
their open research issues. Section VI explains the
relevant design choices made by each routing protocol for
each building block. Finally, section VII concludes the
paper.
II. WMNS: FRAMEWORK, CONSTRAINTS & ADVANTAGES
A WMN (Figure 1) is a self-organized packet-switched
network
composed
of
nodes
with
wireless
communication capabilities and organized in a mesh
topology. The nodes composing WMNs may be
categorized as Wireless Mesh Routers (WMRs) and
Wireless Mesh Clients (WMCs). WMRs are static and
non-power constrained and WMCs are potentially
mobile. WMRs may, in turn, be divided into two types: 1)
WMRs directly connected to the Internet (usually referred
to as gateways), and 2) WMRs used to reach the
gateways or other WMRs by traversing the multi-hop
mesh topology. Throughout this paper, we refer to the
former as gateways and to the latter as WMRs. Due to
space constraints, we will merely focus on the routing
issues between static nodes in WMNs (backbone
WMNs). Mobility management deserves a full survey
paper on its own. For the same reasons, this paper does
not deal with the integration of routing between the wired
Internet and WMNs.
JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010
Furthermore, we are assuming all the WMRs in the

WMN are peers (i.e., they have the same functionalities
and processing capabilities in terms of routing). This is
different from relay-based wireless networks which
involve the addition of lightweight nodes (i.e., relays)
with fewer processing capabilities and functionalities
than WMRs. A proposal in this direction can be found in
[42].
The WMN routing problem may be defined as follows:
given a source WMR (or gateway) and a destination
WMR (or gateway), provide the necessary routing paths
satisfying a number of constraints related to WMNs. A
routing path is a set of WMRs and wireless links
connecting source and destination. The more relevant
constraints a WMN must handle are outlined in
subsection II.A. On the other hand, a WMN does not only
introduce limitations for the design of a high-throughput
routing protocol. The most important advantages are
identified in subsection II.B.
A. Constraints
The constraints of a WMN may be seen as a mixture of
constraints inherited from wired and/or other wireless
networks. In fact, WMNs share properties present in both
wired and wireless networks. In particular, we highlight
some of the most relevant limitations commonly driving
the design of a routing protocol for WMNs. These
relevant constrains are grouped based on whether they are
specifically present in a wireless network environment or
not (i.e., the constraint is present in any generic network
environment).
1) Common to all Networks
Heterogeneous traffic demand. Applications running
over a WMN are similar to those found on the Internet.
Thus, traffic demands can differ from a reliable file
transfer to real-time services, such as voice or video. A
routing protocol must appropriately handle traffic
patterns generated by different applications. Moreover,
different applications and users may have different QoS
requirements.
Undefined number of nodes (WMRs) and
interfaces. The number of WMRs in a WMN may
strongly vary. Additionally, the number of wireless radio
interfaces per WMR may also vary. These, a priory
unknown, characteristics pose additional challenges to the
design of a generic routing protocol for WMNs. Thus, the
routing protocol must support different network sizes and
number of interfaces per WMR. This is unlike in ad-hoc
networks, for which sizes may be smaller and nodes in
general have a single interface, in general.
2) Wireless Networks
Wireless medium instability. Theoretical maximum
rates are only achievable under ideal conditions. Though
WMRs are static, in real-world deployments, the capacity
of wireless links is conditioned by interference, distance,
and fading. This leads to highly variable links, which may
even cause total path degradation.
Scarce wireless bandwidth. A WMN is expected to
provide similar bandwidth to that of the wired Internet.
Several efforts at the physical layer, such as Orthogonal
Frequency Division Multiplexing (OFDM) and Multiple
Input Multiple Output (MIMO) systems have been

developed to provide higher rates. But despite these
recent advances at the physical layer, WMNs still have
bandwidth limitations compared to the wired
infrastructure.
B. Advantages
A WMN can also offer several advantages. The
advantages are a combination of some desirable
properties from wired and wireless networks. Their
combination is, in general, not present in any other
known network environment. The more important
advantages are summarized as follows:
Stable backbone. Unlike in other wireless networks,
WMRs from WMNs are usually static and non-power
constrained. Therefore, a WMR can be augmented with
additional hardware equipment, like storage units and
multiple radios. Moreover, WMR computations are not
energy constrained. Therefore, a stable backbone may
facilitate the design of a routing protocol.
Wireless broadcast medium. In a wireless
environment, the broadcast medium offers the possibility
to transmit a packet to several potential next-hops in
transmission range at the same cost of a unicast
transmission. This provides spatial diversity by exploiting
the fact that the packet is received by multiple receivers.
Multi-rate transmission. A single wireless radio may
offer several choices for the link rate transmission. There
is a trade-off between the number of hops of a path and
the rate selected. Higher rates may lead to increased
number of hops. Lower rates may imply fewer hops, but
lower packet delivery ratio. The effective selection of
wireless link rate by the routing protocol may lead to
much better performance.
Multiple radios. To cope with bandwidth limitations,
a WMR router, as routers in wired networks, is often
equipped with multiple interfaces. Appropriate channel
assignment and antenna selection allow exploiting this
feature in WMNs. Furthermore, a multi-interface WMR
may be potentially equipped with multiple technologydiverse radios, with the consequent heterogeneous
characteristics in terms of coverage and rate. This may
increase the flexibility of the routing protocol, potentially
increasing the routing alternatives.
Figure 1. WMN Architecture.
Multiple channels. An intelligent channel assignment

scheme may substantially improve the throughput of the
213
routing protocol. For instance, if WMRs are equipped

with multiple radios, the multi-channel feature may allow
for concurrent transmissions from the same WMR
without interferences, by assigning an orthogonal channel
to each interface. In this way, the routing protocol may
benefit from multi-channel features to minimize
interference.
III. ROUTING IN WMNS: PROPERTIES & CHOICES
As a consequence of the above constraints of WMNs,
one may extract a list of desirable properties that a
routing protocol should ideally incorporate in its design.
They are presented in subsection III.A. Furthermore,
there are some choices to be made, which are
summarized in subsection III.B, so that these properties
are present in a routing protocol. These choices represent
current approaches found in the literature. As shown in
table I, each potential choice in the design of the routing
protocol provides one or more of the desirable properties.
A. Desirable Properties
To deal with the specific constraints of WMNs, a
routing protocol oriented to maximize throughput for
WMNs should ideally have these main properties:
Bandwidth aggregation. In such a bandwidthconstrained network environment, a routing protocol
must exploit the rich multi-hop wireless mesh topology in
order to support bandwidth aggregation.
Reliability. As WMNs offer multiple paths, the
routing protocol should learn several routes between a
source and a destination in order to react to link or WMR
node failures.
Distributed operation. Routing control information
must be distributed throughout the network.
Low path stretch. The use of the best paths becomes a
necessity. What best means depends on the scenario, but
often, and due to wireless link instability, the best path is
not the shortest path (in number of hops) as in wired
networks. Eventually, a path that takes into account the
wireless constraints whilst not being very different from
the shortest one (i.e. low path stretch) should be provided.
Scalability. A routing protocol must not see its
performance substantially affected by an increase neither
in the number of nodes nor in the number of interfaces
per node.
Low control message overhead. As bandwidth is a
scarce resource, a goal of a routing protocol is to reduce
the overall overhead injected to the network.
Load balancing. When a path is highly congested, all
or part of the traffic may be switched to alternate paths,
so as to efficiently utilize network resources.
Wireless-aware decision-taking. The routing layer
should be aware of the unpredictable events in the
underlying wireless MAC and Physical layers of the
wireless network.
B. Choices
Due to the specific properties that routing protocols
under consideration may ideally have, there are several
strategies that may be incorporated in its design. In other
214
words, there are some major choices to be made when

designing a routing protocol targeting high-throughput
WMNs. After an extensive reading of the existing
literature, we identify the most important approaches that
may be integrated as part of a routing scheme. There are
two mutually exclusive approaches inherited from
classical routing protocols, namely:
Hop-by-hop routing. In Hop-by-hop approaches, all
the WMRs share the responsibility of determining the
route to a destination, as routing decisions are taken at
each hop.
Source routing. In source routing, the source WMR
computes and maintains the entire path to a destination.
When focusing on WMNs, there are several strategies
(not necessarily exclusive) employed in the design of the
routing protocol:
Opportunistic (or anypath) routing. The shared and
broadcast nature of wireless links may be seen as a
feature rather than a limitation to cope with. For instance,
in IEEE 802.11 broadcast mode, a single transmission is
sufficient to potentially be received by several WMRs in
transmission range. Therefore, the probability of
successful reception by any of the WMRs is increased
without increasing the number of total transmissions.
An approach usually associated to opportunistic
routing is network coding. Usually, data has been routed
through the network without being manipulated or
modified. However, with network coding, the transmitted
data may be coded by WMRs and decoded at the
destination. A coding operation is usually a combination
of packets previously received. Thus, each coded packet
contains information about different packets, which
allows, in some cases, to avoid hop-by-hop
retransmissions.
Multi-rate routing. This approach consists of
selecting the more appropriate physical link rate to
maximize throughput gains. There is a trade-off between
reliability (at low link rates) and throughput (at high
rates), which should be optimized.
Overlay routing. Strong similarities exist between
P2P overlays and WMNs, like decentralization and selforganization. Due to these similarities, the wireless
networking research community has tried to push
Distributed Hash Tables (DHTs) from application down
to the network level. With respect to the DHT routing
geometry, multiple paths are often present between
WMRs. The key idea is that underlying DHT geometrical
structures may allow finding a good trade-off between
state stored at each WMR, path stretch, and overhead
introduced in the WMN.
Multi-radio multi-channel routing. The decreasing cost
of wireless devices makes adding multiple interfaces
more feasible. The addition of multiple radios is, in
general, associated to the use of multiple channels in
order to support concurrent transmissions, which
increases the capacity of wireless networks. Channel
assignment techniques may vary the quality of a path due
to changes in the level of interference, and hence also in
traffic load. Moreover, changes in the topology may
generate broken paths. The coordination between routing
and channel assignment becomes essential in multi-radio

multi-channel networks.
TABLE I.
Specific Properties & Choices
Properties
Choices
Wireless-aware
Multi-radio multi-channel, opportunistic
Bandwidth
Network coding, overlay, opportunistic,
aggregation
multi-radio multi-channel, multi-rate
Scalability
Overlay, geographic
Path stretch
Overlay, multi-rate
Distributed
Overlay, geographic
Overhead
Overlay, geographic
Reliability
Opportunistic, network coding, multi-rate
Load balancing
Opportunistic, multi-radio multi-channel
Geographic routing. Scalability may be one of the

most challenging issues to solve in WMNs. Stateless
geographic routing approaches exploit the use of existing
positioning techniques, like GPS, to avoid the
management of routing tables. To route to any
destination, each WMR only relies on position
information of its local vicinity, hence decisions are taken
based on local information.
Additionally, and common to all previous routing
strategies, a component that is always present is some
form of routing metric. Metrics can be classified into
routing metrics and link quality metrics. A link quality
metric quantifies the cost of a link, whilst a routing metric
quantifies the cost of a path.
As already mentioned, due to wireless link variability,
shortest path routing in terms of number of hops is not the
best choice as routing metric. Consequently, routing
metrics that are aware of path quality fluctuations are
needed.
Given the static nature of WMRs, it is feasible to
devote a certain portion of the available wireless
bandwidth or other resources to carry out link (or path)
quality measurements. They may be either passive (i.e. by
using the same data traffic itself) or active (i.e. by
injecting additional control traffic). Moreover, and due to
wireless link quality variation, active or passive
measurements may allow providing the required finegrained path cost estimation.
IV.TAXONOMY
Most WMN routing proposals presented in this section
have a distributed implementation that is used to evaluate
the routing protocol. This means that they are
experimentally evaluated in a real testbed or, at least, they
have a distributed implementation in a network simulator.
Thus, the focus is not on theoretically optimal protocols
that assume a centralized knowledge of the whole
network, but on implementable ones.
After summarizing all potential choices identified in
section III, the goal of this section is to provide a
taxonomy that reflects how the field of WMN research
has used and combined some of these choices. After a
review of the literature, one may notice that not all of
them have been assigned the relevance that would have
justified a new branch in the taxonomy (e.g., geographic
routing). Specifically, the chosen choices are
consequence of some properties present in a WMN. The

taxonomy is presented in Figure 2 and developed
throughout this section.
The proposed classification is based on how each
specific property in a WMN is exploited to maximize
throughput gains. As recently presented in [40], the focus,
in terms of performance metric to optimize, shifts from
connectivity to throughput when moving from ad-hoc to
WMNs. Therefore, there are other wireless routing
protocols in the literature that are out of the scope of this
taxonomy. For instance, those routing protocols solely
focused on maintaining connectivity between nodes. As
they have not been specifically designed for increasing
throughput in WMNs, they do not attain the throughput
level of the routing protocols considered in this paper.
As a result, we have classified routing protocols that
exploit WMN-specific characteristics, and are oriented to
maximize throughput gains into two main approaches
(Figure 2).
The first one relies on the joint use of multiple radios
and multiple channels. This approach may be further
subdivided based on the type of antennas employed,
namely directional or omnidirectional. Furthermore,
routing protocols using omnidirectional antennas have
been further classified as coupled or decoupled depending
on how tight is the relationship and dependencies of
routing and channel assignment.
The second one is based on exploiting the broadcast
nature of the wireless medium. In this approach, the
subset of proposals referred to as opportunistic routing
have been classified into two subgroups that differ in the
relationship between the routing protocol and the
selection of the link rate of the wireless cards. In singlerate approaches, the routing protocol does not select the
link rate, whilst it does in multi-rate approaches.
Furthermore, there is another subset of proposals that also
exploits the additional CPU and storage capabilities a
WMR provides by using one of the two identified
network coding strategies: intra-flow or inter-flow
network coding.
Figure 2. Taxonomy
Previous classifications of routing protocols for

WMNs found in the literature ([29], [30]) categorize
routing protocols for WMNs based on different criteria
215
than the one used in this paper. On the one hand, authors
from [29] classify routing protocols for WMNs based on
an architectural component of the routing protocol. More
specifically, they categorize routing protocols depending
on how they discover and maintain routing paths. On the
other hand, [30] classifies routing protocols based on
their objectives of performance optimization.
Furthermore, the above literature covers a wider
scope of routing, as not all of them were specifically
designed for WMNs. For instance, [29] includes
protocols initially designed for mobile ad-hoc networks.
In [30], routing protocols using the hop count metric are
also studied, which do not fully exploit the advantages of
WMNs. By contrast, our chosen criterion to classify
routing protocols (i.e., what WMN-enabled feature is
exploited to maximize throughput), is different from
criteria previously used in the literature.
A. Multi-Radio Multi-Channel Routing
The coordinated use of multiple radios and multiple
channels per WMR may improve throughput in WMNs.
With an intelligent channel assignment scheme, radios
can also work at the same frequency band, but tuned to
orthogonal channels. The 802.11a, 802.11b/g, and 802.16
standards provide multiple frequency channels, which
may provide an efficient use of the available spectrum
when appropriately configured to orthogonal channels.
As a result, throughput is expected to substantially
increase, which is mainly due to the feasibility of
transmissions occurring in parallel in multi-radio WMRs
and minimization of interference. This is not feasible in a
single radio WMR.
Therefore, routing protocols should ideally work in
cooperation with a channel assignment scheme. The main
goal of a channel assignment strategy is the minimization
of interference. On the other hand, the routing protocol
determines the paths followed by data packets, and hence
the traffic load distribution. In turn, the traffic load
distribution determines the interference. Thus, a channel
assignment strategy that cooperates with a routing
protocol may provide substantial throughput gains.
A WMN offers an ideal architecture for multi-radio
multi-channel routing. First, the non-power constrained
WMN backbone allows adding multiple radio
technologies per WMR. Second, when appropriately
configured to orthogonal channels, the addition of radio
technologies working even in the same frequency band is
no longer an issue. And third, endowing with multiple
radios a WMR is economically feasible due to the
availability of cheap off-the-shelf commodity hardware.
After a careful review of the literature, one may
observe a conceptual difference between proposals that
use omnidirectional antennas and those using directional
antennas.
1) Multi-Channel Routing with Omnidirectional
Antennas
This is the most common approach in the literature.
First, omnidirectional antennas are cheaper. Second, due
to their radiation pattern, in dense topologies, they may
potentially offer increased successful reception
probabilities, due to the number of potential receivers in
216
transmission range. However, for the same reason,

dealing with omnidirectional antennas can also turn out
into increased contention issues without proper medium
access coordination.
We have subdivided this group of routing proposals
into those coupled and those decoupled with the channel
assignment scheme. By coupled channel assignment and
routing solution we refer to proposals in which there is a
tight relationship between routing and channel
assignment. In other words, the routing protocol
determines the channel assignment strategy. On the other
hand, a decoupled scheme is a multi-radio routing
approach that assumes an independent multi-channel
assignment strategy in which the assignment is precomputed.
a) Decoupled Channel Assignment and Routing
(Class A)
The routing protocol itself does not have any influence
on the channel assignment strategy at all. A simple
independent channel assignment strategy to improve
throughput is to configure each radio in a WMR to
different non-interfering channels. This assumption is
made in Multi-Radio Link Quality Source Routing (MRLQSR) [1] and Multi-Radio Ad-hoc On-Demand
Distance Vector (MR-AODV) [3]. Multiple radios are
used to increase the number of candidate paths, hence
offering more paths with potentially more throughput.
They are based on giving higher preference to the more
channel diverse path. Giving higher preference to the
more channel diverse path may result in a decrease of the
contention level between data packets belonging to the
same flow traversing a certain path. Moreover, in [2] and
[3], addressing the interference from other flows also
becomes a concern.
b) Coupled Channel Assignment and Routing
(Class B)
In this case, the routing protocol itself determines the
dynamics of the channel reassignment strategy. In [11],
Raniwala et al. present a routing protocol tightly coupled
with a dynamic channel assignment scheme. At a routing
level, WMRs build a tree-based topology. Trees are
rooted at each gateway of the WMN. WMRs periodically
compute a cost metric aimed at joining the less loaded
tree. The cost metrics evaluated are the hop count to the
gateway, the gateway link capacity, and the gateway path
capacity. As a consequence, they may periodically switch
between multiple trees, hence providing load balancing
among the gateways.
On the other hand, and concerning the channel
assignment strategy, each WMR periodically exchanges
its channel load within its interference range. Channel
usage is estimated by using the number of interfering
WMRs and the aggregated traffic load contributed by
each WMR. Then, a WMR selects the less loaded channel
not currently assigned to a WMR in a higher hierarchy
level in a routing tree. It is assumed that traffic load
grows as a WMR is higher in the tree hierarchy.
Therefore, when assigning channels, a high level WMR
has priority over low level WMRs.
As the routing protocol dictates the traffic load

distribution and traffic load dynamically changes the
channel assignment strategy, the dynamic load-aware
channel assignment scheme turns out to be driven by the
routing protocol. Furthermore, the tree hierarchy imposed
by the routing protocol constrains the channel assignment
scheme by determining the channel assignment priorities
of WMRs.
Reference [37] presents a Topology and Interference
aware Channel assignment architecture (TIC), which
describes a hybrid channel assignment technique working
in cooperation with a routing protocol based on the
Weighted Cumulative Expected Transmission Time
(WCETT) [1] metric and the Dijkstra algorithm for
minimum path cost computation. It is assumed that each
WMR has one radio tuned to the same channel in the
same physical layer technology. Furthermore, all these
radios have a unique and static channel assigned. The rest
of the wireless radios are dynamically assigned.
To assign a channel to each radio, each WMR builds a
conflict graph in order to assign orthogonal channels. A
conflict graph is used to represent interference between
wireless links. Channel assignment and route quality
evaluation is done in parallel by combining the use of
Dijkstra with the conflict graph. For a source WMR, each
non-interfering channel not present in the conflict graph
is checked to compute the minimum cost next-hop using
WCETT. From all the possible channels and neighbor
WMRs, those which minimize the routing cost are
chosen. Once the next-hop and channel are determined,
the Dijkstra search advances one hop to the intended
destination and computes the minimum cost and channel,
and so on. Therefore, in this particular case, the channel
assignment strategy is determined by the routing
protocol.
Additionally, when a WMR changes the channel of
one of its radios, data traffic transmitted through this
channel switches to the common radio interface.
Therefore, the common channel network is used as
default network for routing flows until channel
assignment in other channels ends. In this sense, the
channel assignment scheme also influences routing
operation.
2) Multi-Channel Routing with Directional Antennas
(Class C)
Directional antennas may significantly simplify the
routing protocol. A wireless link composed of two
directional antennas is similar in concept to a wired link.
In a wired link, there is no interference and it may
potentially offer long transmission ranges. In a wireless
link composed of directional antennas, especially when
there are not many available channels, directional
antennas may offer spatial separation to handle
contention. Therefore, interference between antennas may
be minimized, or totally suppressed. Moreover,
directional antennas offer increased transmission range
by decreasing the number of hops to reach the destination
WMRs. In turn, this decreases the number of medium
access contentions.
However, the widespread deployment of directional

antennas poses several challenges. On the one hand,
automatically-steerable
directional
antennas
are
expensive. On the other hand, if they are not steerable,
the direction to which the antenna points must be changed
manually, which may also be costly.
But this problem is smaller in WMNs, as they have a
static WMN backbone. All WMRs and their associated
antennas are fixed; hence cheap directional antennas may
be located by pointing to a known predefined neighbor.
Routing and channel assignment with cheap directional
antennas has been addressed in Directional Optimized
Link State Routing (DOLSR) [31]. The WMN is
composed of a tree rooted at the gateway. A radio with an
omnidirectional antenna, configured to the same channel
in every WMR in the network, is used to collect control
information. The control information is used to decide
which WMR could be the parent in the tree of the WMR.
Information gathered from the omnidirectional antenna is
used to select the more suitable directional parent. Thus,
the directional antennas are used only for data packet
transmission. They also evaluate different channel
assignment strategies, decoupled from the routing
protocol, which is similar to the approaches in [1]-[3].
Reference [33] studies the WMN routing problem
when directional and omnidirectional antennas coexist.
Essentially, Dynamic Source Routing (DSR) computation
is used with a modified version of the Expected
Transmission Count (ETX) [41] metric. When a WMR
does not have a route to the intended destination, it floods
the network by sending control packets that carry the
accumulated ETX value. The ETX computation method
is modified so that they not only discover neighbors in
the omnidirectional range but also in the directional
range. Basically, different values of transmission power
covering omnidirectional and directional ranges are
tested.
B.Opportunistic Routing
Opportunistic routing is based on the following
principle: when a WMR wants to transmit a data packet,
instead of transmitting it in unicast mode (i.e., to a single
next-hop), it directly broadcasts the data packet. The
broadcast transmission permits the senders to not
necessarily know which WMR is the next-hop.
Afterwards, the routing protocol decides on-the-fly which
of the potential receivers of the broadcasted packet may
forward the data packet, and thus become the next-hop.
The potential receivers of the packets need to work in a
coordinated way in order to minimize forwarding of
duplicated packets. In turn, the forwarding of data
packets is also done in broadcast mode. The coordination
process entails the need for specific opportunistic routing
metrics and mechanisms to decide the best receivers.
Therefore, in opportunistic routing, the next-hop is
known after data packet transmission, which is contrary
to classical unicast routing approaches. In classical
unicast routing approaches, the next-hop is known before
the data packet is forwarded. WMNs are a suitable
candidate for incorporating the opportunistic routing
philosophy. This is because with a dense and static
217
backbone, the number of potential receivers of a

broadcast packet increases. Thus, opportunistic routing
may provide robustness in the transmission. Specifically,
we will focus on opportunistic routing protocols that
effectively increase throughput perceived by the
destination WMR.
The static WMN backbone offers inherent path
diversity. Path diversity is provided due to the existence
of a rich mesh topology that also offers point to
multipoint links. Thus, opportunistic routing protocols
can exploit the point-to-multipoint transmissions a WMN
provides in order to maximize throughput gains.
Opportunistic routing tackles a known problem in
WMNs, namely short-term path quality variations. An
issue in WMNs is to guarantee high throughput paths due
to the high variability of the link quality. In general,
classical routing protocols are not able to update link
costs (and thus not able to update path costs) at the fine
time scale wireless link variations occur. They usually recompute wireless link costs at the scale of various
seconds. For instance, the unicast ETX [41] metric is
recomputed every 10 seconds. Path recalculation is done
at an even coarser time scale. In Opportunistic routing,
each packet may potentially follow a different path. Each
data packet takes a single path but none of the successive
packets are forced to follow this path. There are some
possible candidate paths but none of them is chosen a
priori. In fact, the path is chosen on-the-fly depending on
the current, usually point-to-multipoint, link status.
Furthermore, as transmission is done in a broadcast
manner, the transmission rate is significantly improved.
This is because, in 802.11, for each data packet
transmitted, there is the exponential backoff delay
mechanism. For unicast transmissions, this may lead to
excessive delays when handling retransmissions.
Moreover, even when the transmission is successfully
received, the data packet needs to be acknowledged, not
allowing the potential sender to transmit a new data
packet until the ACK control packet is received. On the
other hand, in broadcast mode, there are neither per-link
retransmissions nor acknowledgment procedures. Thus,
WMRs may potentially transmit at higher rates, being
only limited by the physical carrier sensing. As a
consequence, throughput may be increased.
We classify opportunistic routing protocols into two
different categories: single-rate and multi-rate. In singlerate opportunistic routing, it is assumed that the WMRs
are not able to manage the data rate at which packets are
transmitted. On the other hand, in an opportunistic multirate environment, the routing protocol selects both the
forwarding next-hop and the data rate for each radio in
the WMR.
1) Single Rate Opportunistic Routing (Class D)
The main issue in these protocols is how to decide
which neighbors forward data packets. The main
challenge that arises is the cooperation between the
potential next-hop set in order to select the best
forwarding WMR, where best often means the WMR that
maximizes throughput gains.
218
Extremely Opportunistic Routing (ExOR) [16]

computes at each potential forwarder the shortest path to
the intended destination. The shortest path is estimated by
summing up the link costs associated to the path
calculated using the Dijkstra algorithm. The link costs are
computed using the ETX [41] metric. Thus, each WMR
has all the ETX values to reach the destination and it
calculates the minimum ETX value. For coordination
purposes, a forwarder priority list is sent in each data
packet to schedule the order of forwarding attempts by
the next-hop set. As a result, a WMR only forwards a
data packet if all higher priority WMRs failed to do so.
In Resilient Opportunistic Mesh Routing (ROMER)
[15], the key idea is that each packet carries a credit
which is initially set by the source and is reduced as the
packet traverses the network. As in ExOR, each WMR
also computes a path cost for forwarding a packet from
itself to the intended destination.
In ROMER, a data packet may be duplicated when
traversing the WMN. This may happen because potential
next-hops may forward data packets if the credit of the
packet is high enough. The credit associated to each data
packet is decremented at each forwarding step according
to the WMR credit cost, which basically means that more
credits are consumed as the packet moves away from the
shortest path to the destination; hence data packets are not
forwarded through these paths.
2) Multi-Rate Opportunistic Routing (Class E)
Leveraging rate control to select the optimal rate in
opportunistic routing may lead to throughput gains.
Specifically, some broadcast links may be underutilized,
hence losing throughput. A potential improvement driven
by the routing protocol consists of increasing the data rate
to increase throughput, and hence the optimal utilization.
(Note that this is independent from MAC layer auto-rate
algorithms based on unicast MAC layer procedures, not
compatible with opportunistic routing, such as counting
the number of retries per packet to determine the optimal
rate.)
On the other hand, as higher transmission rates entail
shorter radio ranges, link loss rate may potentially be
increased. Therefore, the network could eventually
become disconnected. A solution for these links is to
decrease the link rate of the WMR, thus increasing the
number of potential next-hops (i.e., increasing
connectivity).
Nevertheless, achieving the optimal rate poses several
challenges. To date, there are some recent proposals
addressing multi-rate. First, as different rates mean
different transmission ranges, there is a trade-off between
the rate selected and the number of hops. Choosing a high
rate may decrease reliability, thus requiring more data
packet (re-)transmissions. A low bit rate may guarantee
reliability, but it may also result in an unnecessary
decrease of throughput. Therefore, depending on the bit
rate selected, the set of potential next-hops of a WMR is
variable.
Zeng et al. [24] propose Multirate Geographic
Opportunistic Routing (MGOR), a heuristic for
opportunistic multi-rate routing, which takes into account
the constraints imposed by transmission conflicts. They

argue that the problem is NP-hard, and so heuristics are
used to find a solution. Specifically, two different
heuristics are proposed. One is Expected Advancement
Rate (EAR), which addresses what next-hop is closer to
the destination in distance by using location information.
The other heuristic is Expected Medium Time (EMT),
which is based on a generalization of ETT [1].
On the other hand, Shortest Multirate Anypath
Forwaring (SMAF) [19] computes the opportunistic
multi-rate path by modifying a generalized version of the
Dijkstra algorithm. The main contribution consists of
reducing the number of neighbor combinations to test for
finding the optimal next-hop neighbor set at a given rate.
Specifically, the optimization is based on testing a
number of combinations equal to the number of
neighbors, and not all the possible combinations for each
tested rate, which is an exponential number of
combinations. This is the key to achieve a polynomial
algorithm, hence reducing computational complexity.
C. Opportunistic Network Coding Routing
In the reviewed routing schemes, network coding is an
add-on to opportunistic routing. But it is classified as a
different group in the taxonomy, due to the qualitative
conceptual change that exploiting the CPU and storage of
WMRs may entail.
Essentially, in opportunistic network coding routing,
WMRs mix the content of data packets. Then, at each
hop, they transmit in a broadcast manner the resulting
coded packet over the point-to-multipoint wireless
medium. Therefore, every coded data packet received at
an intended destination contains information about
different original packets. Different received data packets
contain information of some original packets, thus
providing, in general, useful incremental information to
the receiver. Additionally, network coding may employ,
if needed, original non-previously coded packets received
at the destination.
Luckily, network coding poses specific requirements
that may be easily fulfilled by WMN backbone nodes.
First, WMRs need to keep remarkable state information
to store data packets. Second, for some network coding
scenarios, each WMR is recommended to be highly static
in order to facilitate the buffering of data packets to be
combined. Finally, WMRs require considerable CPU
operations for mixing packets. As a consequence, WMRs
should not be power-constrained. We have grouped
network coding routing into two main groups: intra-flow
network coding and inter-flow network coding.
1) Intra-Flow Network Coding (Class F)
Intra-flow network coding is based on mixing packets
belonging to the same data flow. This is, in fact, a
specific case of single-rate opportunistic routing. When a
source wants to send data packets to a destination, the
source WMR breaks up the file into batches of packets
and keeps transmitting packets in broadcast mode from
the same batch until the batch is acknowledged by the
destination. However, in intra-flow network coding,
there is no coordination between the receivers of data
packets.
Before forwarding data packets, the forwarders store

them in a buffer. When enough data packets are stored,
the forwarder computes a random linear combination of
the packets. The mixed packets are headed to the same
destination. The randomness in the mixing procedure
assures with high probability that different WMRs will
not forward exactly the same packets. Thus, the number
of packets received by the destination is increased. It is
likely that WMRs participating in the forwarding
procedure send different combinations of packets.
In MAC-independent Opportunistic Routing &
Encoding (MORE) [14], every WMR sends probe
packets to capture the link costs associated to its
neighbors. Upon link calculation, the cost of each link is
flooded to the whole network so that the Dijkstra
algorithm can calculate the shortest paths. The cost
associated to each link is calculated by means of the ETX
[41] metric. The WMRs check whether they are closer to
the destination than the transmitter or not by using
Dijkstra combined with ETX. If this is the case, they
store the received coded packets in a buffer. When a
forwarder has sufficient data packets, it makes a random
linear combination of received data packets, thus
generating new coded data packets, and it eventually
forwards the coded packets. This process continues at
each hop until enough data packets are received by the
destination so that it is able to decode the original
information. In order to decode the original data packets,
a common constraint for the receiver is that the number
of innovative coded data packets received must be greater
or equal to the number of original data packets.
Furthermore, to support reliability, the destination WMR
sends and ACK (using unicast best path routing based on
ETX) to the source when it has received enough coded
data packets.
Gkantsidis et al. present Multipath Code Casting
(MCC) in [39], which also employs intra-flow network
coding. However, in this scheme, link costs are collected
and propagated by an overlay routing discovery module
called Virtual Ring Routing (VRR) [13]. Moreover, a
credit-based distributed algorithm is used for rate control.
And, backpressure theory ideas are used in order to avoid
congestion and keep the length of each WMR queue
bounded.
2) Inter-Flow Network Coding (Class G)
In inter-flow network coding, the coding operation is
done over data packets belonging to different data flows.
Coding Opportunistically (COPE) [16] is based on
mixing packets generated by different flows when a
WMR detects an advantage for doing this operation. An
advantage is usually detected when the number of coded
packets transmitted in a single transmission may be
maximized and the destination has enough information to
decode the packet. To detect an advantage, a WMR has to
gather some information of the flows present in the
network. In COPE, Dijkstra and ETX [41] are used for
computing minimum cost paths.
Distributed Coding-Aware Routing (DCAR) [35] goes
beyond COPE and suggests combining the route
discovery process with the detection of coding
219
opportunities in order to maximize the inter-flow network

coding opportunities. On-demand source computation
combined with the ETT metric are used for calculating
minimum cost paths. The basic idea of this scheme is to
discover intersecting paths, instead of choosing disjoint
paths for certain flows in the network. These flows are
such that making them coinciding in a WMR to code
their packets is beneficial for achieving network
throughput gains. Thus, end-to-end throughput of
different flows is maximized.
V. DESIGN BUILDING BLOCKS
The functionality required by a high-throughput
routing protocol may be split into three building blocks,
namely neighbor discovery, control message propagation,
and route determination. These building blocks are
common to the broad set of network protocols studied in
the previous section.
Essentially, every building block addresses one
specific function, which is a part of the routing protocol.
First, WMRs have to gather link cost information about
its neighbors (i.e., neighbor discovery).
Second, information about link costs must be distributed
throughout the network to the appropriate WMRs,
implying a certain propagation of route control messages
over the WMN. This is handled by the control message
propagation building block. Finally, once the necessary
routing information is collected by all parties, the routing
paths to the destination nodes are determined (i.e., route
determination).
Notice also that the characteristics of WMNs highly
influence the strategy adopted by each building block to
carry out its function. Throughout this section, the
different underlying strategies and their inter-dependency
with the exploited characteristics are studied for each
identified building block. In brief, in neighbor discovery,
the radiation pattern of the antennas; in control message
propagation, the stable non-power constrained backbone;
and in route determination, the forwarding approach
employed and the possibility of employing multi-rate
features. Moreover, open issues related to each building
block are also discussed.
A. Neighbor Discovery
The neighbor discovery building block groups the
functionality related to the process of determining which
WMRs can be reached by means of direct communication
(i.e., without having to cross any other intermediate
WMR). Periodic or non-periodic broadcast packets are
usually used to discover the WMRs reachable by direct
communication. This could be sufficient to maintain a
neighbor table in each WMR if wireless links were as
stable as wired links. However, in wireless links, the
neighboring relationship is mainly determined by the
quality of the link. As it is highly variable and unstable,
the wireless link quality is not limited to the same two
classical states as in wired networks. In classical wired
networks, it is usually assumed that a link works well or
does not work at all. Moreover, wireless link quality may
vary depending on the direction of the link, which results
220
in wireless link asymmetry. In practice, this means that

the neighbor discovery building block is in charge not
only of discovering WMRs in the physical proximity, but
also of estimating the quality and stability of a link
towards each WMR within transmission range.
Depending on this latter estimation, a neighboring
relationship will be established or not.
WMNs offer an environment that enables accurate
wireless link estimation. A basic approach would send a
control packet and then wait for an answer. However, the
non-power constrained nature of WMNs allows
implementing more elaborate and complex procedures to
increase the wireless link cost estimation accuracy. On
the other hand, in WMNs, the radiation pattern of the
antennas equipping WMRs may cause considerable
changes in the issues tackled to measure link quality.
Therefore, link quality measurement procedures and
their associated link quality metrics, taking into account
the antenna radiation pattern, become key issues of the
neighbor discovery building block. Link quality metrics
are explained in the first subsection. The second
subsection summarizes the different procedures to carry
out depending on the antenna radiation pattern.
1) Link Quality Metrics
The cost associated to each link, which will be later
used to calculate the routes in the route determination
building block, requires the computation of link quality
metrics in the neighbor discovery building block. Note
that link metrics are conceptually different from routing
metrics. As mentioned in section II, link quality metrics
quantify the cost associated to a wireless link, and they
are handled by the neighbor discovery process. On the
other hand, routing metrics are handled by the route
determination building block, as they measure the quality
of paths, and not merely of single links. Thus, routing
metrics are built by using link metrics as input to quantify
the cost of an end-to-end route path.
As shown in table II, for each of the proposals found in
the literature, two main factors condition the design of
wireless link metrics: the primary metric employed and
the measurement technique utilized to calculate the
parameters to estimate. Besides, the procedure may vary
depending on the radiation pattern of the antenna. Finally,
the link cost to be used by the route determination
building block (referred to as metric in table II) is also
presented.
Primary metric. A primary metric is an indicator used
to quantify the quality that a wireless link includes. There
are four primary metrics [12] used in the literature,
namely:
Packet delivery ratio (PDR): The more common
parameter chosen in the literature (e.g., [1]- [5], [37], and
[41]) quantifies wireless link reliability at a packet level.
The PDR is the ratio of packets correctly
received/captured to the total number of packets sent by
the sender. The PDR is usually calculated in both
directions of a wireless link in order to deal with link
asymmetry, which is common in wireless links.
Bit error rate (BER): This is the ratio of bits with
errors to the total number of bits that have been received
during a given time period. The BER primary metric

defines the reliability at a bit level.
Signal-to-interference plus noise ratio (SINR): The
extent to which the power of the received signal exceeds
the sum of noise plus interference at the receiver. SINR
quantifies the quality of the received signal.
Received signal strength indication (RSSI): This is the
signal strength observed at the receivers antenna during
packet reception. RSSI defines the quality of the signal
received.
Measurement technique. The most common
measurement technique used to measure packet delivery
ratio is based on the probe packet concept.
Probe packet: It consists of periodically broadcasting
or unicasting a packet of fixed size. The packet contains
the number of probe packets received by the sender.
Therefore, the receiver of the probe packet can calculate
the delivery ratio of the link in the receiver-to-sender
direction.
Packet pair: Packet Pairs are a special case of probe
packets. In ETT [1], Metric of Interference and ChannelSwitching (MIC) [2], and Interference-Aware Metric
(iAWARE) [3], a WMR sends two unicast probe packets
of different size. The receiver WMR measures the
difference between the instants in which each packet is
received, and it forwards this information to the sender.
Then, it is used to estimate the available bandwidth of a
link.
Furthermore, based on the strategy employed to
generate probing packets, the approaches followed may
be categorized as active, passive and cooperative.
Active: A WMR explicitly sends control packets to
discover its neighbors. This is the default procedure in
most proposals explored, either sending probe packets or
packet pairs.
Passive: In Efficient and Accurate link-quality
monitoR (EAR) [5], discovery can be made with the use
of data packets. The real traffic generated in the network
is also used as probing packets without incurring extraoverhead.
Cooperative: In [5], a WMR overhears data packets
transmitted by each of its neighbors to estimate the link
quality from its neighbors to itself.
Proposal
ETX[41]
ETT[1]
mETX[4]
EAR[5]
PowerETX[33]
TABLE II.
Link Quality Estimators
Primary
Measurement
Antenna
metric
technique
Omni/dir
PDR
Probe packet
Omni/dir
PDR
Packet pair
Omni/dir BER,PDR
Probe packet
Probe packet
Omni/dir
PDR
passive
cooperative
Omni/dir
PDR
active
MIC[2]
Omni
PDR
Packet pair
iAWARE[3]
Omni
SNR/SINR
PDR
Packet pair
ETP[38]
Omni
PDR
Probe packet
Link Quality
Estimates
Loss rate
Bandwidth
Loss rate
Bandwidth
Loss rate
Bandwidth
Interference
Bandwidth
Interference
Bandwidth
Interference
Link Quality Estimates. The final goal of the

measurement procedure is to quantify the link cost by
means of one or more link quality estimates, which are
obtained by appropriately combining one or more
wireless link primary metrics. The metrics found in the
literature reviewed follow:
Loss Rate: Most of the proposals try to measure the
loss rate which is the percentage of packet/bit losses in
the link. The loss rate is usually measured by means of
probe packets, which are used to calculate the PDR
primary metric. The values obtained for the PDR primary
metric are used to compute the packet loss rate. For
instance, these values may be averaged by means of an
exponentially weighted moving average (EWMA). In
[33], a modification of the ETX metric (Power-ETX) is
proposed to deal with a WMN composed by WMRs
equipped with omnidirectional and directional antennas.
It is based on alternating the transmission of broadcast
probe packets at two different transmission power levels.
Each of these transmission powers covers the
omnidirectional and directional range. As a result, a
WMR may discover neighbors that are beyond the
omnidirectional range, which are neighbors in the
directional range.
On the other hand, modified ETX (mETX) [4] takes
into account the average and standard deviation of the
BER primary metric of the captured packets to calculate
loss rate. The standard deviation may potentially be
useful in order to quantify wireless link variability.
Bandwidth: There are proposals ([1]-[5], and [38])
focused on measuring the available bandwidth of the
wireless link. Available bandwidth is usually captured
through the use of packet pairs.
Interference: Interference caused by neighbors of a
WMR in transmission range, may also be a parameter to
estimate. Specifically, the interference measured is
interflow-interference. This is interference generated to
each other by/to packets belonging to different flows.
Essentially, to measure the interflow-interference ([2],
[3], and [38]), monitoring methods are employed to
capture the number of interfering WMRs at each wireless
link. In general, this estimate is associated to the use of
omnidirectional antennas.
Measurement
techniques
introduced
by
omnidirectional antennas are based on sensing the
medium and exchanging the captured information. For
instance, in [2], a rough estimation is made to count the
number of interfering neighbors of a WMR. On the other
hand, [3] uses the measured SNR and SINR primary
metrics to capture inter-flow interference variations.
2) Dependency of Neighbor Discovery on the
radiation pattern of the antennas
The procedures followed to perform neighbor
discovery vary depending on the radiation pattern of the
antennas equipping WMRs. According to their radiation
pattern, antennas can be classified into directional and
omnidirectional.
a) Omnidirectional Antennas
An omnidirectional antenna has a uniform radiation
pattern in all directions. The discovery of neighbors with
221
omnidirectional antennas becomes straightforward with

the use of broadcast probe packets [41].
On the other hand, interference requires special
attention, as studied in [1]-[3], and [38]. For instance, in
high-density WMNs, where WMRs are equipped with
omnidirectional antennas, contention should be carefully
handled due to the potentially high number of
neighboring WMRs.
b) Directional Antennas
Neighbor discovery with directional antennas is more
challenging, since it introduces additional issues with
respect to neighbor discovery with omnidirectional
antennas. With directional antennas, WMRs must
appropriately manage the direction and beam width of the
antennas in order to maintain their neighbors. On the
other hand, non-steerable antennas must be manually
installed.
When direction and beam width of the antennas are
correctly managed, the deafness problem may be
minimized. Deafness occurs when a transmitter is unable
to communicate with its intended receiver, because the
antenna of the receiver is not pointing to the transmitter.
To discover the neighbors of a WMR, some approaches
have been proposed:
Probabilistic discovery: In [32], probe packets are sent
in a random direction and beam width, to calculate the
direction and beam width of the antenna.
Omnidirectional neighbor discovery: As stated in [31],
omnidirectional antennas could be used to handle control
messages for directional antennas. Specifically,
omnidirectional antennas are used to discover the
neighbors of directional antennas and estimate their
associated wireless link costs.
Link quality metrics with directional antennas may be
simplified with respect to the link quality metrics for
omnidirectional antennas. Directional antennas increase
spatial separation for contending transmissions compared
to contention in omnidirectional antennas. Therefore,
estimates that quantify interference may not be necessary,
as presented in table II.
An appropriate channel
assignment scheme offering frequency separation may be
sufficient to deal with contention in scenarios with
directional antennas. Therefore, as shown in table II, link
quality interference estimates may not be totally
necessary. In principle, proposals, such as [41], may be
directly employed with directional antennas. However,
due to their different physical layer properties, link
metrics should be slightly modified, as detailed in
previous subsection [33].
OPEN RESEACH ISSUES
A list of the identified open research issues related to

the neighbor discovery building block follows:
Link quality. Vlavianos et al. [12] suggest that every
single primary metric on its own may not be a good
estimate of link quality. A proof of this fact in an indoor
testbed may be found in [12]. These studies showed that
although BER may be a good predictor, it requires a high
number of computations to make the appropriate
222
measurements. On the other hand, RSSI cannot capture

interference and SINR is quite complex to be measured.
A starting point may be a deep review of the effects
captured by each of the different primary metrics by
studying appropriate combinations of the primary metrics
in order to find an accurate link quality metric.
Control overhead vs network resources. When
designing a new wireless link metric proposal, there is a
trade-off between the control overhead devoted to
measure link quality with enough accuracy and the
network resources consumed. Most current work is
devoted to send control packets that contend with data
packets, and thus affect the behavior of the network. This
may be solved by applying passive measurement
strategies. However, an in-depth analysis providing
accuracy bounds of these techniques may be required.
Active measurement strategies. Regarding active
measurements, the schemes followed by recent literature
are quite similar. Active measurement techniques are
based on periodically sending broadcast or unicast probe
packets. An effort should be made to study other
measurement strategies. For instance, in general, the size
of the probe packet and the inter-generation time of probe
packets are fixed. A future research direction may consist
of evaluating whether changes in the active measurement
strategy may lead to more accurate link quality metrics.
Self-interference: Active probe packets have the
disadvantage of affecting the wireless link quality they
are measuring. Current wireless link metrics do not take
into account the interference generated by the active
measurements. A detailed study of the impact of
interference caused by active measurements may be of
interest.
Wireless link quality prediction. A parameter not
sufficiently evaluated in current work on wireless link
quality assessment is how to predict the variability of a
wireless link. Keeping historical measurements or storing
traffic patterns to predict the future state of a link may
provide a starting point.
Link quality estimation with directional antennas.
The spatial separation offered by directional antennas is
able to decrease the complexity on the calculation of a
link metric. However, directional antennas have their own
physical layer properties. A subject of further study may
be the definition of estimators specific to directional
antennas rather than using those originally designed for
omnidirectional antennas.
B. Control Message Propagation
The control message propagation building block is
responsible for sending all the necessary routing control
messages to the appropriate WMRs.
The cost incurred by the transmission of control
messages is not as critical as in power-constrained ad-hoc
networks. As a consequence, an appropriate control
message propagation building block for WMNs should be
aimed at maximizing overall throughput by propagating
accurate routing information of wireless link
measurements, even at the expense of being more costly.
On the other hand, a secondary goal is to decrease the
overall overhead incurred by the building block. In this
sense, as WMNs offer a stable backbone, no additional

control messages due to the movement of nodes must be
sent.
In WMNs, the literature distinguishes between two
different traffic pattern scenarios: 1) traffic only
exchanged between WMRs and gateways, and 2) traffic
exchanged between any pair of WMRs. Throughout this
section, we refer to the former as any-to-gateway and the
later as any-to-any. References [11], [31], and [38]
assume the any-to-gateway scenario. References [6]-[10],
[13], and [26] assume the any-to-any scenario.
Furthermore, in WMNs, both traffic scenarios have in
common the use of some strategy to perform the
propagation of route control messages. Depending on the
particular protocol, control message propagation could be
carried out by one (or both) of two components: route
dissemination and route discovery. Their common goal is
to provide the necessary route control information to the
route determination building block. And the potential
coexistence of route discovery with route dissemination is
facilitated by the stability of the WMN backbone. This is
explained in the first subsection. On the other hand,
depending on the combination between route
dissemination and discovery, different techniques are
presented in the literature to efficiently propagate control
messages. They are discussed in the second subsection.
1) Route Dissemination & Route Discovery
The goal of route dissemination and route discovery
approaches is the same, i.e., obtaining the necessary
routing information from the network to compute the
routes, but the way in which they obtain such information
is different.
Route dissemination refers to the process of
propagating information about link state previously
obtained by the neighbor discovery building block. And,
this information is periodically disseminated to the
network in a proactive way, i.e., without any WMR
asking for it. There are some key design decisions to
make, such as the accuracy of the information to
disseminate. For instance, in [6] the accuracy of the
information disseminated is decreased as the distance in
hops from the disseminator to the recipient WMR
increases.
In addition, there is another method for obtaining
routing information from WMRs in a WMN, which is the
route discovery process. It is triggered by a source WMR
for obtaining the necessary routing information ondemand. Therefore, it is done in a reactive way, that is,
when the source WMR has data packets to send to a
certain destination. In brief, this process usually works by
sending control messages that asks for route information
to the WMRs they traverse. Once these control messages
obtain the requested routing information, they are sent
back to the requesting WMR.
In WMNs, the route dissemination and route discovery
components may need to work in cooperation. In other
words, both components may complement each other. For
instance, in large WMNs, route discovery may help route
dissemination to complete the propagation of required
route control information, as it is not practical in this case
that all WMRs keep state about the rest of the WMRs in
the network. Therefore, the existence of route discovery
may be highly dependent on the procedure followed for
route dissemination and the other way around.
The stability of the WMN backbone facilitates the
coexistence of both route dissemination and discovery
components. For instance, one may employ a set of well
known (i.e., by all WMRs) static WMRs to which all
routing information is disseminated. As this set of WMRs
is not mobile, it facilitates any WMR requester to
locate/access them.
Nevertheless, in some cases, route dissemination may
be sufficient to obtain the necessary routes. For instance,
in small WMNs, a flooding-based dissemination scheme
may be appropriate. In this case, each WMR in the
network has enough information to route packets to any
destination without incurring into excessive overhead due
to the small size of the WMN. Furthermore, depending on
certain WMN requirements (e.g., delay), the route
discovery process may be sufficient to obtain the desired
routes.
2) Techniques for Propagating Control Messages
Route dissemination and route discovery require a
massive transmission of control messages throughout the
network. Therefore, it is fundamental that this is done as
efficiently as possible. In this subsection, we present a
brief review of representative methods for propagating
routing control messages.
Every routing protocol may have an associated
technique for propagating useful control messages over
the network. Table III presents a summary of this section.
We have categorized control message propagation
schemes as tree-based, efficient flooding, and all-to-some
propagation. Moreover, the components (dissemination
and/or discovery) used to gather routing information for
each studied proposal are also presented. Finally, the
traffic pattern scenario assumed by each proposal is also
shown.
A discussion of each of the propagation schemes
follows.
Tree-based: Several approaches in the literature are
based on tree topologies ([11], and [38]). Such a tree
structure is used in any-to-gateway scenarios.
Essentially, the root of the tree is a gateway in the
WMN. Thus, as many trees as gateways are built. These
trees are usually built in an incremental way, i.e., they are
expanded as WMRs join the network. If there are
multiple trees, a recently joined WMR must decide which
tree to join.
The construction and maintenance of a tree topology
determines specific control message propagation
strategies. In MAC-Aware Load Balancing (MaLB) [38],
each WMR disseminates the accumulated routing
information to its parent WMR. Specifically, each WMR
propagates to its parent the cumulative routing
information of all the WMRs for which it is root of the
subtree that includes all WMRs from leaf WMRs up to
itself. On the other hand, in [11], the gateway
disseminates its routing information to the rest of the
WMRs in the tree following the tree-like structure.
223
Moreover, there is a route discovery component that

requests to the gateway, which is the root of the tree, if a
new WMR is allowed to join the tree. Then, the gateway
sends an answer to the new WMR.
TABLE III.
Approaches to Send Control Messages
Proposal
Scenario
Components
Strategy
Clustering[26] Any-to-any
Dissemination Efficient flooding
Fisheye[6]
Any-to-any
LOLS[7]
Any-to-any
OLSR[8]
Any-to-any
Dissemination Efficient-flooding
Gossip[17]
Any-to-any
Dissemination Efficient-flooding
Dissemination
ORRP[9]
Any-to-any
All-to-some
Discovery
Dissemination
VRR[13]
Any-to-any
All-to-some
Discovery
Dissemination
Hyancinth[11] Any-to-gw
Tree
Discovery
MaLB[38]
Any-to-gw
Dissemination
Tree
In both above approaches, the control message

propagation is influenced by the tree topology.
Specifically, the propagation of control messages in treebased proposals is such that leaf WMRs do not forward
control messages.
Efficient flooding: Flooding is a well-known technique
to propagate messages to all the WMRs in the network. A
WMR disseminates (i.e., route dissemination component)
a message to all its neighbors and these neighbors, in
turn, transmit to all of its neighbors, and so on, until all
the WMRs receive the message. However, this may incur
in unnecessary duplicated transmission of packets. To
avoid reception of duplicated packets, [8] proposes to
use the multipoint relay scheme. This strategy is based on
acquiring 2-hop neighbor information in order to select
the minimum number of 1-hop neighbors that guarantee
successful reception of all 2-hop neighbors.
On the other hand, the scheme presented in [17],
follows a gossip-based approach, in which each receiver
decides with a certain probability if the control message
is forwarded or not. Essentially, a source WMR sends a
control messages with probability one. A WMR forwards
a control message with probability p and discards the
control message with probability 1-p. If a WMR receives
a previously received control packet again, it is discarded.
Although not specifically studied for WMNs,
clustering [26] could potentially be another strategy
employed to reduce the overhead caused by route control
messages. Clustering is based on partitioning the network
into groups of WMRs called clusters. The forwarding of
control messages is limited to cluster heads and cluster
gateways. A cluster head is chosen so that all WMRs in
the cluster receive control messages. A certain WMR is
elected as cluster gateway to forward route control
messages to other clusters. In this scheme, additional
route control messages to elect the cluster head as well as
the gateways must be sent to build the clusters.
A specific case in efficient flooding approaches is
partial-flooding approaches. In [6], the flooding process
only covers a certain area of the network close to the
source WMR. Besides, the flooding may be done at
different frequencies depending on the range covered. In
224
fact, there are some strategies focused on reducing the

frequency of flooded messages. One approach proposed
in [6] is to define different frequencies of transmissions
depending on the distance in number of hops from the
disseminator WMR. Thus, the more distance, the less
frequently routing information is disseminated.
Adidtionally, the disseminated information may vary
depending on the dissemination period. In Localized Ondemand Link State (LOLS) [7], for short periods, each
WMR sends route control information which quantifies
average values of link costs. This route information is
sent to the entire network. On the other hand, for longer
periods, WMRs disseminate route control information
which quantifies current link cost to WMRs in the
neighborhood.
All-to-some: In an all-to-some approach, all WMRs
maintain routing entries to some WMRs. The routing
information stored in each WMR differs. The goal is to
keep routes to a sufficient number of WMRs in the WMN
so that it is guaranteed that any intended destination is
reachable. These proposals pose several advantages. For
instance, there is no flooding process involved. And
opposed to flooding-based approaches, route state
information stored at each WMR is substantially reduced.
To obtain the routing information, some proposals based
on sending the requests in some strategically predetermined directions have been conceived ([9], and
[13]).
A representative example of all-to-some schemes is
VRR [13]. In fact, VRR explores the idea of porting
overlay routing concepts, usually used at Application
layer, to sit directly above the MAC layer. An overlay is
basically a routing structure that relies in an underlying
network routing protocol. Specifically, VRR employs a
ring-like structure. Every WMR maintains paths only for
their virtual neighbors, which are some predecessors and
successors in the ring structure. Virtual neighbors may be
separated in the physical network, hence requiring multihop paths to reach each other. Thus, a virtual hop may be
composed of multiple physical hops. VRR exploits this
dichotomy to assure that following the virtual paths is
sufficient to reach any intended WMR. To build these
virtual paths, a request (route discovery) is sent to find
the WMRs that are virtually closest to the requester in the
ring structure. Furthermore, the request includes routing
information about the requester (route dissemination) in
order to update routing information at the requested
WMRs.
In Orthogonal Rendezvous Routing Protocol (ORRP)
[9], a request (route discovery) is sent in orthogonal
directions until it finds a WMR with the route
information requested. Furthermore, each WMR
periodically disseminates (route dissemination) its routing
information in two orthogonal directions. Thus, the
number of control messages in the network is decreased
compared to a flooding approach by sending control
messages to only two orthogonal directions. This strategy
is based on the idea that two pairs of orthogonal lines
intersect in a plane.
OPEN RESEARCH ISSUES
A list of the identified open research issues related to

the control message propagation building block follows:
Propagated information. Given the instability of
wireless links, a challenge that arises is what exact
information is going to be spread to other WMRs in the
WMN. For instance, one possible option is to associate a
predicted lifetime to the propagated wireless link cost.
This may be calculated by the neighbor discovery
building block.
Intelligent dissemination. Future work should also
target the minimization of the routing overhead. For
instance, one may consider an intelligent strategy that
only disseminates routing control messages when
relevant changes occur. A question that arises here is
what is considered as a relevant change in a WMN. There
is a proposal for generic networks (i.e., not only WMNs)
suggesting this possibility in [10].
Route dissemination vs route discovery. There is no
generic agreement in the procedure to carry out to
propagate route control messages. Although route
discovery may exist, it is not clear its relative importance
with respect to route dissemination. The route
dissemination component may yield lower delays but
considerable overheads costs. On the other hand, the
route discovery component may lead to higher delays but
lower overhead costs. Probably, the trade-off between
delay and overhead may depend on the WMN
requirements which may dynamically change. As a result,
the importance between route dissemination and
discovery may ideally vary over time. Therefore, the
introduction of mechanisms devoted to gather dynamic
WMN requirements so that the relation between these
components is optimal may be a subject of further study.
Paths with enough available bandwidth. In general,
the studied routing protocols globally assume that there is
a path with enough available bandwidth to the
destination. However, when there is not enough
bandwidth available to later send the data packets, path
discovery should be avoided. In WMNs, it is usually
assumed that there is a path between any pair of WMRs.
But even though there is a path, it is not guaranteed that
there is sufficient available bandwidth to maintain a
communication. The introduction of route discovery or
dissemination mechanisms that provide path reservation
may be of potential interest. For instance, this may be
done within the control message propagation building
block by marking reserved nodes, hence avoiding their
participation in other potential paths.
Overlays. A promising approach for efficient control
message propagation is the use of overlays for
propagating routes. However, there is a primary challenge
to face. In fact, porting P2P overlay routing systems to
the network layer is not trivial, as there are some
differences to take into account. First, pushing
Distributed Hash Tables (DHTs) on top of the link layer
makes connectivity between any couple of WMRs
become an issue. And second, the mapping of logical
paths of P2P structures into physical paths does not take
into account the underlying physical topology, which
leads to path inefficiencies. In general, this is not a

problem in wired networks, due to their higher link rates.
However, this is no longer true in WMNs where
bandwidth is a scarce resource. Consequently, the study
of strategies able to generate logical paths that are similar
to physical paths may be of potential interest. For
instance, this may potentially be done by setting up some
rules to apply when a WMR joins a WMN, so that its
assigned location maintains the logical structure without
compromising the path stretch of the WMN.
C. Route Determination
Based on the routing information (e.g., link cost
information) gathered by means of the control message
propagation building block, the route determination
building block is in charge of determining the most
appropriate routing paths from a certain WMR to any
other WMR (any-to-any scenarios) or from/to the
gateway (any-to-gateway scenarios). Thus, the expected
outcome of the route determination building block is the
computation of routing tables that specify the next-hop
for incoming data packets. Furthermore, to compute
routing tables in WMNs, it is necessary to take into
account the different methods incoming data packets may
be forwarded.
As a result, the route determination building block
depends on two main components, namely forwarding
approach and route computation. The particular
properties of a WMN allow a WMR to forward data
packets using different approaches (i.e., unicast and
broadcast), which are explained in the first subsection.
The different forwarding approaches have several
implications in the route computation design, which is
introduced in the second subsection. Specifically, the
forwarding approach has several implications on the
algorithms employed to compute the routing tables.
These algorithms determine the path that has the
minimum route cost metric to the intended destination.
Additionally, the routing metric design depends on the
utilized forwarding approach. (A routing metric is used to
quantify the cost of the paths to the intended destination.)
1) Forwarding Approach
In WMNs, there are two methods for forwarding data
traffic through the network to the next-hop. On the one
side, there exists the option of deterministically
unicasting the data packet from one WMR to one of its
neighbors, which is selected by looking up the precomputed routing table. On the other side, one may
broadcast from one WMR to all WMRs in transmission
range. Therefore, in a broadcast forwarding approach,
various WMRs may potentially be the simultaneous
receivers of a data packet.
The unicast approach handles the wireless link in the
same way forwarding in wired networks does, i.e., as if it
was a point-to-point wired link. A directional or
omnidirectional antenna may potentially be used in
WMRs following the unicast approach. However, for
environments where direction of data packets is known
and unique, it may be more efficient to associate a
directional antenna with the unicast forwarding approach.
225
On the other hand, the broadcast approach, changes the

classical concept of link. In a shared wireless medium,
transmission matches a point-to-multipoint distribution
rather than a point-to-point one. Thus, omnidirectional
antennas are specially suited to exploit such kind of links,
where may be useful to send data packet in all possible
directions and/or received by multiple next-hops.
2) Route Computation
A static and non power-constrained WMR can perform
costly route computations which is not feasible for nodes
belonging to power-constrained wireless networks.
Therefore, a WMR may, in general, use shortest-path
algorithms to calculate the more appropriate routing paths
without taking into account battery or CPU-load issues.
The algorithms to compute the minimum path cost to
the intended destination can be categorized as Dijkstra,
Bellman-Ford, and local-based.
Dijkstra: In link state routing approaches, the link
costs of the entire network are disseminated by using an
adequate strategy to propagate control messages to the
intended receiver. In this approach, the algorithm to
compute the shortest path commonly used in WMNs is
based on modifications of the well-known Dijkstra
algorithm.
Bellman-Ford: In Bellman-Ford-based routing
protocols, routes are computed in a more distributed
manner. In this case, a WMR receives information about
the network after being processed by its neighbors. The
distance-vector approach is used. There are various
flavors of the Bellman-Ford algorithm. For instance, one
of these flavors is used in on-demand source routing to
update the path cost carried in the control packet at each
hop [35].
Local-based: The calculation is done in a greedy
manner, which selects the best next-hop closer in distance
to the destination by only using local information ([15],
[20], and [13]). It is calculated on a hop-by-hop basis
during data transmission.
Depending on the size of the network, Dijkstra and
Bellman-Ford algorithm require costly CPU operations
and considerable storage capabilities. Furthermore, the
cost associated to the algorithms may change with the
forwarding approach. Specifically, in a broadcast multirate forwarding approach, as there are various potential
next-hop neighbor and rate choices, the number
operations to carry out minimum path cost may increase
[19]. On the other hand, as only local information is
handled to compute the next-hop, local-based route
computation tends to consume less considerable CPU.
The minimum path cost algorithm comes together with
a routing metric. As mentioned in section II, the path cost
is quantified by means of a routing metric, as opposed to
the link cost, which is quantified by a link quality metric.
The computation of a routing metric takes as input
parameter a set of link costs calculated during the
neighbor discovery process. A subset of these link costs
will form part of the resulting minimum cost path. The
cost represented by the routing metric in use may be
calculated by means of three different methods. First, it
may only use local information to compute the cost.
226
Second, it may be the sum of the weights of the cost of all

links in the path.
(Recall that each link cost was previously calculated by
the neighbor discovery building block.) Finally,
additional information may be required to compute a
more elaborated function.
Figure 3. Taxonomy of Routing Metrics Issues
There are two different philosophies to compute the

routing metric depending on the strategy followed in the
forwarding approach component. They are categorized as
unicast and broadcast routing metrics.
a) Unicast Routing metrics
Although any kind of antenna may potentially be used
with protocols employing unicast routing metrics, these
metrics are probably more suited for directional antennas.
One may group routing metrics into those not aware
and those aware of the operation of the MAC protocol
(see Figure 3). Focusing on 802.11 networks, some
considerations follow for each of the groups.
Non 802.11-aware unicast routing metrics. In this
case, the link metric does not directly take into account
either link contention or channel usage or interference.
Shortest path algorithms using as metric the sum of the
weights of link metrics, like ETX [41] and ETT [1] are
some representative proposals.
802.11-aware unicast routing metrics. Other metrics
to calculate the optimal route are aware of the operation
of the MAC layer. This routing metrics take into account
the variation of wireless link quality in the route
determination building block. Basically, the factors that
are relevant to determine the quality of a path, and are
captured by 802.11-aware routing metrics are the
following ones:
Intraflow-interference: The interference due to packets
belonging to the same flow. A common method to
measure the intraflow-interference is to estimate how
channel diverse are the links composing a path. As
showed in table IV, intra-flow interference is captured in
references ([1]-[3], and [38]).
Interflow-interference: This type of interference is
usually calculated during link quality estimation by the
neighbor discovery building block. References [3] and
[38], as showed in table IV, capture inter-flow
interference.
Literature on routing metrics often presents weighted
average functions of different measured components ([1][3]), such as intra-flow or inter-flow interference. The
link cost metric is modified by modeling some
interference level which may be measured in different

ways. A common feature in such approaches is that the
calculated routing metric reflects the cost of the path from
one single WMR to the intended destination.
On the other hand, [38] evaluates how the overall
network performance would be degraded if a new WMR
joins one of the different trees. Eventually, the WMR
joins the tree that minimizes the global delay associated
to transmitting a bit for all the WMRs in the forest, which
is a union of trees rooted at the gateways, to its associated
gateway. As a result, load balancing is explicitly
provided. Therefore, the routing metric takes into account
what would happen to the overall network quality when a
new path is chosen.
TABLE IV.
MAC-Aware Unicast Routing metrics
Routing metric
Interference
WCETT[1]
Intra-flow
iAWARE[3]
Intra/inter-flow
MIC[2]
Intra-flow
MaLB[38]
Intra/inter-flow
b) Broadcast Routing metrics

This operation requires at least one static broadcast
wireless interface per WMR. Additionally, WMRs are
static and may be equipped with an omnidirectional
antenna. Consequently, broadcast traffic distribution
perfectly fits WMNs. In fact, broadcast routing metrics
are usually associated to opportunistic routing protocols.
Though routing metrics belonging to unicast
forwarding have been employed in broadcast forwarding
[16], these routing metrics are not totally appropriate for
these environments because they do not take into account
all the potential path opportunities [22].
First, there is the issue of finding the optimal set of
neighbors that guarantees maximum advancement to the
intended destination, which is not present in unicast
forwarding. This is similar in concept to the neighbor
discovery building block. However, in this case, the
discovery of the neighbors is dependent on the current
intended destination.
Second, another singular issue is that of prioritizing
neighbors amongst those in the selected neighbor set. The
optimal candidate neighbor set is the union of neighbors
in transmission range for a WMR that maximizes
progress to the destination. Additionally, there exists a
trade-off between the number of neighbors available (to
maximize reception probability) and the number of
neighbors that truly add some progress as next-hop. Thus,
broadcast routing metrics are based on finding the
optimal candidate neighbor set that adds more progress
towards the destination. As shown in table V, one may
classify the neighbor set selection and prioritization into
two groups: single-rate and multi-rate (see also Figure 3).
Single-rate neighbor set selection and prioritization.
Expected Any-path Transmissions (EAX) [21] and
Remaining Path Cost (RPC) [22] metrics try to select and
prioritize the number of forwarding candidates from all
those belonging to the neighbor set of a WMR. As
showed in table V, these proposals are based on
generalizing the well-known ETX [41] metric to account

for the expected number of anypath transmissions. The
expected number of anypath transmissions is the
estimated number of broadcast transmissions so that the
intended destination could eventually receive a data
packet.
Routing metric
EAX[21]
RPC[22]
EAR[20]
EMT[24]
EATT[19]
TABLE V.
Broadcast Routing metrics
Rate
MAC coordination
single
yes
single
no
multi
yes
multi
no
multi
no
Based on
ETX
ETX
Location info
ETT
ETT
In [22], all the possible neighbor WMR combinations

at each potential next-hop in the path towards the
intended destination are collected. After that, the optimal
candidate forwarder is selected by using a generalization
of Bellman-Ford algorithm.
Multi-rate neighbor set selection and prioritization.
As showed in table V, location information has also been
used to decide which WMRs of the potential neighbor set
will forward the data packet [20]. And this selection is
done by combining location information with an
appropriate tuning of the underlying transmission rate,
hence exploiting the underlying multi-rate transmission
capabilities, to provide a heuristic for maximizing
advancement to the destination at each hop.
In Expected Anypath Transmission Time (EATT) [19],
wireless link quality is measured by checking the possible
rates achievable. Furthermore, an algorithm to compute
optimal routes based on Dijkstra is proposed. As for the
metric, ETT [1] (see table V) metric is generalized to
account for the multiple rates in an anypath environment.
On the other hand, in [24], a generalization of the ETT
metric is proposed as broadcast routing metric. The
proposed candidate selection and prioritization may be
computed using a distributed Bellman-Ford algorithm.
Finally, besides guaranteeing that an optimal neighbor
set selection choice is made, the broadcast routing metric
used must minimize the number of duplicated packets as
well as packet contention. There are some common
generic techniques to solve this issue. Some of them are
based on introducing a scheduler algorithm aware of the
wireless medium, while others exploit network coding,
hence not needing additional coordination at all (Figure
3). As a result, the coordination strategy may be classified
into two groups: 802.11-aware coordination and non802.11-aware coordination (see table V).
802.11-aware coordination. References [16], [20],
and [21] note that sending data packets in broadcast mode
requires a scheduler to avoid duplicated transmissions by
the potential forwarders. Thus, a coordination strategy to
avoid (or minimize) this situation must be in place. In
general, such a strategy is based on using timers
associated to the MAC layer [20]. And it works as
follows. Every data packet carries the WMR forwarding
priorities calculated in the sending WMR. And these
priorities are calculated by exploiting the available
location information to determine the distance to the
227
destination. Furthermore, the MAC broadcast layer is

modified to transmit an ACK packet when a data packet
is received. Then, a WMR candidate forwarder with jth
priority order may wait for the time needed for
transmitting j-1 ACK packets before deciding to send if it
does not overhead any previous ACK.
Furthermore, in [16] and [21], control messages are
exchanged between the forwarding WMRs to schedule in
order their forwarding attempts. Therefore, a WMR
forwards a data packet only if higher priority WMRs
failed to do so.
Non 802.11-aware coordination. With network
coding, each receiver mixes received packets before
forwarding them. Random network coding assures each
receiver will not forward the same packets. Packets
belonging to the same ([14], and [39]) or different ([34],
and [35]) flows may be combined. The main advantage is
that no explicit coordination between WMRs is needed
because the probability that two WMRs use the same
linear combination is quite low. Therefore, random
network coding exploits spatial diversity and increases
throughput due to the absence of such an explicit
coordination scheme.
A list of the identified research topics related to the

route determination building block that may need further
work follows:
Interference estimation. A known problem faced by
unicast routing metrics is how to quantify the two types
of interference in the form of a routing metric. For
instance, MIC [2] , iAWARE [3] and MaLB [38] capture
intra-flow and inter-flow interference, but each proposal
requires different methods for interference estimation.
Therefore, there is no consensus in the research
community in how to measure interference in a WMN.
Specifically, it is somewhat unclear whether
information from lower layers (i.e., Physical and MAC)
may be necessary to obtain accurate interference
estimations. Therefore, a research direction may consist
of measuring interference merely using the network level
without resorting to lower layers. Furthermore, an
evaluation of its achieved accuracy to see the necessity of
using cross-layer interactions may be required.
Integration of the routing metric with the rest of
components. The routing metric designed may not work
properly with any routing computation algorithm. The
design of the routing metric is tightly coupled with the
design of the rest of the components of a routing protocol.
For instance, it is shown in [2] that the WCETT [1]
metric combined with the Dijkstra algorithm does not
provide isotonicity, where isotonicity means that a
routing metric should ensure that the weighted order of
two paths is preserved if they are appended or prefixed by
a common third path. Thus, WCETT cannot be calculated
locally for each WMR and then simply perform a
summation to obtain the cost of the whole path. In other
words, WCETT requires a single calculation with the
presence of all the WMR components involved in the
path quality calculated by WCETT, namely the ETT of
228
each link and the channel assigned to each link.

Otherwise, the calculated routing path may be nonoptimal or even may trigger routing loops. Therefore, the
design of accurate isotonic routing metrics may be of
interest.
Route recalculation timers. There is no consensus on
appropriate values of the expiration timer that triggers the
recalculation of the quality of a route. There is a trade-off
between the optimal route choice and the stability of the
route [23]. Frequent route path changes may lead to
packets not received in order at the receiver. Thus,
routing pathologies may occur at high scale. Some
metrics may be needed to decide whether it is an
advantage to change the routing path on a per-packet
basis as anypath routing may assume.
Dijkstra and Bellman-Ford requirements. The
overall overhead required for a minimum path calculation
using algorithms such as Dijkstra or Bellman Ford is not
scalable as network size increases. In fact, all WMRs
must be aware of the link costs of the whole network.
And this is not feasible in a large-scale WMN, even
assuming that WMRs may embed powerful processors.
Approaches to minimize or restrict such requirements to
certain areas should be investigated.
Broadcast transmission limitations. The broadcast
forwarding approach introduces one major issue, namely
the absence of a reliability mechanism similar to that
present in unicast forwarding. This may imply shifting
reliability mechanisms to the routing level, i.e.,
guaranteeing reliability hop-by-hop, by areas, or in an
end-to-end basis.
VI.TAXONOMY VS BUILDING BLOCKS
This section summarizes and qualitatively compares
the most relevant features of each the routing protocols
considered in this survey. Previous sections have
highlighted the operation of their building blocks. In this
section, we highlight and summarize the most relevant
design decisions each routing protocol made for each of
its building blocks (see table VI). Each representative
routing proposal is tagged with a letter (A, B, C, D, E, F,
G) identifying the classes defined in section IV.
Furthermore, for each building block, we identify the
more important aspects out of those discussed in section
V.
As for neighbor discovery, we focus on the link quality
metric as the more relevant aspect. As shown in table VI,
the ETX [41] metric seems to be the most common
approach employed in the literature. Some other relevant
proposals choose the ETT metric [1], or even, not to
estimate the link quality at all and merely use hellos to
discover the neighbors. Another relevant link metric is
iAWARE [3], which is used by MR-AODV [3].
As for control message propagation, table VI presents
the propagation technique implemented by each routing
proposal. Flooding is the most common approach
followed by the generic routing approaches explored. But
other relevant alternatives exist. For instance, an all-tosome approach is implemented in MCC [39], a tree-based
approach in Hyancinth [11], and an efficient-flooding
approach in DOLSR [31]. Another interesting

observation is that in ROMER [15] and MGOR [24],
control message propagation is not needed due to the
specific operational characteristics of these protocols.
As for route determination, we take into account three
main features. The first one is the interaction with the
MAC layer when computing the routing metric
(rightmost column in table VI). Furthermore, in case such
interaction is present is based on two operational
principles: 1) the MAC coordination (see the route
determination building block), implemented by ExOR
[16], MGOR [20], and DCAR [35], and 2) whether the
routing metric takes into account the MAC layer,
implemented by MR-LQSR [1], MR-AODV [3], and TIC
[37]. The second feature examined for route
determination, is the algorithm employed for minimum
cost path computation, namely Dijkstra, Bellman-Ford,
and local-based. As shown in table VI, the most common
strategy followed by routing protocols shown is Dijkstra
algorithm. This algorithm is used by MR-LQSR [1], MRAODV [3], MORE [14], ExOR [16], SMAF [19], COPE
[34] and TIC [37]. The Bellman-Ford algorithm is
implemented by DSR [33], and DCAR [35]. And, the
computation of the routes merely using local information
is implemented by ROMER [15], and MGOR [20].
The third feature compared is the forwarding approach,
chosen for transmitting data packets, namely unicast or
broadcast. The unicast forwarding approach is used by
MR-LQSR [1], MR-AODV [3], Hyacinth [11], TIC [37],
DOLSR [31], and DSR [33]. On the other hand,
broadcast forwarding is implemented by MORE [14],
ROMER [15], ExOR [16], SMAF [19], MGOR [20],
COPE [34], and DCAR [35].
TABLE VI.
Taxonomy vs Building Blocks Comparison.
Route Determination
Control
Neighbor
message Forwarding
C Proposal
Route
discovery
MAC
propagation approach computation
A
MRETT
Flooding
Unicast
Dijkstra
Yes
LQSR[1]
MRA
iAWARE Flooding
Unicast
Dijkstra
Yes
AODV[3]
B Hyacinth[11] Hellos
Tree
Unicast
Local
No
B TIC[37]
ETT
Flooding
Unicast
Dijkstra
Yes
Efficient
C DOLSR[31] Hellos
Unicast
Dijkstra
No
Flooding
C DSR[33]
ETX
Flooding
Unicast
Bellman
No
D ExOR[16]
ETX
Flooding
Broadcast
Dijkstra
Yes
D ROMER[15] ETX
No
Broadcast
Local
No
E MORE[14]
ETX
Flooding
Broadcast
Dijkstra
No
E MCC[39]
ETT All-to-some Broadcast
Local
No
F SMAF[19]
ETT
Flooding
Broadcast
Dijkstra
No
F MGOR[20] Hellos
No
Broadcast
Local
Yes
G COPE[34]
ETX
Flooding
Broadcast
Dijkstra
No
G DCAR[35]
ETX
Flooding
Broadcast Bellman
Yes
As shown in table VI, there are not routing proposals

that effectively and simultaneously take into account the
most advances strategies found in the literature for all
building blocks. For instance, for the neighbor discovery
building block, there is some literature explicitly focused
on improving the estimations of link quality. However,
some of these relevant improvements ([4], [5]) are far
from being integrated into a single routing protocol

jointly with other recent improvements in other building
blocks. In other words, existent proposals take into
account one issue belonging to one specific building
block, and they focus on improvements without deeply
studying the potential implications in the other building
blocks of the routing protocol.
A list of the identified generic routing open research

issues that go beyond a single building block follows:
Routing in Cognitive WMNs (CWMNs). A CWMN
is a WMN composed by mesh nodes equipped with
cognitive radios. A cognitive radio is intended to sense
the medium so that spectrum usage is maximized, hence
incrementing the throughput transferred by the network.
Such spectrum usage maximization implies learning the
periods during which cognitive radios may use the
available spectrum (i.e., the periods when users on
licensed bands do not use the spectrum).
The introduction of cognitive radios adds several new
open research issues to the routing procedures, especially
when the availability of free spectrum bands is quite
intermittent. Little research has been done in this
direction, one recent proposal may be found in [42]. An
open research issue is the synchronization between
WMRs to common spectrum bands to perform
opportunistic forwarding. Particularly, the potential
solutions for this issue may have some similarities with
approaches explored in the field of Delay Tolerant
Networks (DTNs). However, in DTNs the forwarding
opportunities are created based on nodes physical
movement and not spectrum bands dynamicity.
Honest WMR behavior. An open issue is how to
guarantee honest behavior of WMRs. WMRs may behave
selfishly, hence not contributing with their own resources
to the routing of certain flows. This selfish behavior may
lead to reduced throughput in the WMN. An incentivebased routing scheme that stimulates WMRs to route data
packets belonging to other communication flows may be
of interest.
Joint network coding and multi-rate transmission.
Though the optimal multi-rate path was found in [19],
network coding is not used to achieve maximum
throughput. A proposal that combines network coding
with multi-rate may outperform current multi-rate
transmission and network coding approaches in terms of
throughput when considered in isolation.
Joint network coding and multi-radio multichannel WMN features. The combination of network
coding and multi-radio multi-channel approaches may be
explored. In fact, random network coding approaches
found in the literature are designed for single-radio
single-channel static WMNs. In a static non-powerconstrained environment, there is not such a limitation. A
proposal that effectively and jointly exploits multi-radio
multi-channel operation and network coding may
outperform current routing protocol results.
Network coding impact on routing. It is not clear the
impact of network coding in the routing protocol. For
229
instance, further work is needed to determine in what

cases it is convenient to exploit coding opportunities, and
also which flows should be combined.
Therefore, a research direction may consist of a
evaluating the potential modifications to apply to a
routing protocol augmented with network coding may do
to maximize routing performance. Specifically, one
immediate open question is whether it is necessary to
add/update more building blocks to a routing protocol.
VII.CONCLUSIONS
To the best of our knowledge, this is the first survey
focusing on routing protocols that really exploit the
specific features of WMNs. These specific features may
be summarized as the existence of a stable non-powerconstrained backbone.
Therefore, and compared to other wireless networks,
the main goal of such protocols may be shifted from
connectivity to throughput maximization. In this paper,
we provide a complete classification of routing protocols
based on the way they exploit WMN-specific features
(section IV). All current routing protocols for WMNs
aiming at maximizing throughput may be mapped to one
of the classes defined in the proposed taxonomy, which
has two main branches. In the first one, we classify
routing protocols that benefit from the addition of
multiple radios configured in multiple channels to nonpower-constrained WMN nodes. The second one includes
routing protocols that benefit from the stability of the
backbone along with the broadcast nature of the medium
and, in some cases, the processing capabilities of the
nodes.
Furthermore, we identify three common building
blocks found in a high-throughput routing scheme for
WMNs, namely neighbor discovery, control message
propagation, and route determination (section V). Recent
advances for each specific building block, as well as the
design issues they may face in WMNs are also presented.
Some insights on how to improve current routing
approaches are also given. Specifically, for each building
block, a detailed discussion of future open research issues
to solve is provided. Finally, we mapped several relevant
routing proposals present in the taxonomy to the most
significant and common design choices chosen for each
building block (section VI), thus linking the taxonomy
and the building blocks explained in the previous section.
ACKNOWLEDGMENTS
This work was supported in part by the Ministry of
Science and Innovation under grant number TEC200806826/TEC (project ARTICO). Additionally, it was also
partially funded by grant PTQ08-01006439.
REFERENCES
[1] Draves, R., Padhye, J., and Zill, B. 2004. Routing in multiradio, multi-hop wireless mesh networks. In Proceedings
of the 10th Annual international Conference on Mobile
Computing and Networking (Philadelphia, PA, USA,
230
September 26 - October 01, 2004). MobiCom '04. ACM,

New York, NY, 114-128.
[2] Y. Yang, J. Wang, and R. Kravets. Designing routing
metrics for mesh networks. Proceedings of the IEEE
Workshop on Wireless Mesh Networks (WiMesh). IEEE
Press, 2005.
[3] Subramanian, A.P.; Buddhikot, M.M.; Miller, S.,
"Interference aware routing in multi-radio wireless mesh
networks," Wireless Mesh Networks, 2006. WiMesh 2006.
2nd IEEE Workshop on , vol., no., pp.55-63, 25-28 Sept.
2006
[4] Koksal C. E. and Balakrishnan H., "Quality Aware Routing
in Time-Varying Wireless Networks," IEEE Journal on
Selected Areas of Communication Special Issue on MultiHop Wireless Mesh Networks, Volume 24, Issue 11, Nov.
2006 Page:1984 - 1994
[5] Kim, K. and Shin, K. G. 2006. On accurate measurement
of link quality in multi-hop wireless mesh networks. In
Proceedings of the 12th Annual international Conference
on Mobile Computing and Networking (Los Angeles, CA,
USA, September 23 - 29, 2006). MobiCom '06. ACM,
New York, NY, 38-49.
[6] M. Gerla, X. Hong, and G. Pei, Fisheye state routing
protocol for ad-hoc networks, IETF Internet Draft, June
2002, draft-ietf-manet-fsr-03.txt.
[7] Nelakuditi, S.; Sanghwan Lee; Yinzhe Yu; Junling Wang;
Zifei Zhong; Guor-Huar Lu; Zhi-Li Zhang, "Blacklistaided forwarding in static multi-hop wireless networks,"
Sensor and Ad-hoc Communications and Networks, 2005.
IEEE SECON 2005. 2005 Second Annual IEEE
Communications Society Conference on , vol., no., pp.
252-262, 26-29 Sept., 2005
[8] T. Clausen and P. Jacquet, Eds. 2003 Optimized Link State
Routing Protocol (Olsr). RFC. RFC Editor
[9] Cheng, B., Yuksel, M., and Kalyanaraman, S. 2006.
Orthogonal Rendezvous Routing Protocol for Wireless
Mesh Networks. In Proceedings of the 2006 IEEE
international Conference on Network Protocols (November
12 - 15, 2006). ICNP. IEEE Computer Society,
Washington, DC, 106-115
[10] Levchenko, K., Voelker, G. M., Paturi, R., and Savage, S.
2008. XL: an efficient network routing algorithm.
SIGCOMM Comput. Commun. Rev. 38, 4 (Oct. 2008), 1526.
[11] A. Raniwala, Chiueh Tzi-cker, Architecture and
algorithms for an IEEE 802.11-based multi-channel
wireless mesh network, in proceedings of the 24th IEEE
Conference on Computer Communications (INFOCOM),
pp. 2223-2234 March 2005.
[12] Vlavianos, Angelos; Law, Lap Kong; Broustis, Ioannis;
Krishnamurthy, Srikanth V.; Faloutsos, Michalis,
"Assessing link quality in IEEE 802.11 Wireless Networks:
Which is the right metric?," Personal, Indoor and Mobile
Radio Communications, 2008. PIMRC 2008. IEEE 19th
International Symposium on , vol., no., pp.1-6, 15-18 Sept.
2008
[13] Caesar, M., Castro, M., Nightingale, E. B., O'Shea, G., and
Rowstron, A. 2006. Virtual ring routing: network routing
inspired by DHTs. In Proceedings of the 2006 Conference
on Applications, Technologies, Architectures, and
Protocols For Computer Communications (Pisa, Italy,
September 11 - 15, 2006). SIGCOMM '06. ACM, New
York, NY, 351-362.
[14] Chachulski, S., Jennings, M., Katti, S., and Katabi, D.
2007. Trading structure for randomness in wireless

opportunistic routing. SIGCOMM Comput. Commun. Rev.
37, 4 (Oct. 2007), 169-180.
[15] Y. Yuan, H. Yang, S.H.Y. Wong, S. Lu, W. Arbaugh,
ROMER: resilient opportunistic mesh routing for wireless
mesh networks, in: Proceedings of the First IEEE
Workshop on Wireless Mesh Networks (WIMESH'05),
Santa Clara, CA, September 2005.
[16] Biswas, S. and Morris, R. 2005. ExOR: opportunistic
multi-hop routing for wireless networks. In Proceedings of
the 2005 Conference on Applications Technologies,
Architectures,
and
Protocols
For
Computer
Communications (Philadelphia, Pennsylvania, USA,
August 22-26, 2005.) SIGCOMM05. ACM, New York,
NY, 133-144.
[17] Haas, Z.J.;Halpern, J.Y;Li Li,Gossip-based ad-hoc
routing, Networking, IEE/ACM Transactions on ,
vol.14,no.3,pp.479-491.
[18] R. G. Ogier, F. L. Templin, B. Bellur, M. G. Lewis
Topology Broadcast Based on Reverse-Path Forwarding
(TBRPF), Internet Draft, draft-ietf-manet-tbrpf-03.txt,
Nov. 2001.
[19] Multirate Anypath Routing in Wireless Mesh Networks.
Rafael Laufer (University of California at Los Angeles,
US); Leonard Kleinrock (University of California, Los
Angeles, US). In press for IEEE Infocom 2009.
[20] Zeng, Kai; Lou, Wenjing; Zhang, Yanchao, "Multi-Rate
Geographic Opportunistic Routing in Wireless Ad-hoc
Networks," Military Communications Conference, 2007.
MILCOM 2007. IEEE , vol., no., pp.1-7, 29-31 Oct. 2007
[21] Zifei Zhong; Nelakuditi, S., "On the Efficacy of
Opportunistic Routing," Sensor, Mesh and Ad-hoc
Communications and Networks, 2007. SECON '07. 4th
Annual IEEE Communications Society Conference on ,
vol., no., pp.441-450, 18-21 June 2007
[22] H. Dubois-Ferriere, M. Grossglauser, and M. Vetterli,
Least-Cost Opportunistic Routing, in Proceedings of the
2007 Allerton Conference,Monticello, IL, USA, Sep. 2007.
[23] K. Ramachandran, I. Sheriff, E. Belding Royer, and K.
Almeroth. Routing stability in static wireless mesh
networks. In Proc. of PAM, 2007
[24] Kai Zeng; Wenjing Lou; Hongqiang Zhai, "Capacity of
opportunistic routing in multi-rate and multi-hop wireless
networks," Wireless Communications, IEEE Transactions
on , vol.7, no.12, pp.5118-5128, December 2008
[25] Nandiraju, N.; Nandiraju, D.; Santhanam, L.; Bing He;
Junfang Wang; Agrawal, D.P., "Wireless Mesh Networks:
Current Challenges and Future Directions of Web-In-TheSky," Wireless Communications, IEEE , vol.14, no.4,
pp.79-89, August 2007
[26] Kwon, T. J. and Gerla, M. 2002. Efficient flooding with
Passive Clustering (PC) in ad-hoc networks. SIGCOMM
Comput. Commun. Rev. 32, 1 (Jan. 2002), 44-56.
[27] Belair Networks. www.belairnetworks.com
[28] Strix Systems. www.strixsystems.com
[29] M. Campista, P. Esposito, I. Moraes, L. Costa, O. Duarte,
D. Passos, C. de Albuquerque, D. Saade, and M.
Rubinstein, Routing metrics and protocols for wireless
mesh networks, Network, IEEE, vol. 22, no. 1, pp. 612,
Jan./Feb. 2008.
[30] Ian Akyildiz, Xudong Wang. Wireless Mesh Networks,
John Wiley & Sons, 2008.
[31] Das, S.M.; Pucha, H.; Koutsonikolas, D.; Hu, Y.C.;
Peroulis, D., "DMesh: Incorporating Practical Directional
Antennas in Multichannel Wireless Mesh Networks,"
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
Selected Areas in Communications, IEEE Journal on ,

vol.24, no.11, pp.2028-2039, Nov. 2006
Vasudevan, S.; Kurose, J.; Towsley, D., "On neighbor
discovery in wireless networks with directional antennas,"
INFOCOM 2005. 24th Annual Joint Conference of the
IEEE Computer and Communications Societies.
Proceedings IEEE , vol.4, no., pp. 2502-2512 vol. 4, 13-17
March 2005
Cheekiralla, S.; Engels, D.W., "Routing in Heterogeneous
Wireless Ad-hoc Networks," Computer Communications
and Networks, 2007. ICCCN 2007. Proceedings of 16th
International Conference on , vol., no., pp.951-956, 13-16
Aug. 2007
Katti, S.; Rahul, H.; Wenjun Hu; Katabi, D.; Medard, M.;
Crowcroft, J., "XORs in the Air: Practical Wireless
Network Coding," Networking, IEEE/ACM Transactions
on , vol.16, no.3, pp.497-510, June 2008
Jilin Le; Lui, J.; Dah Ming Chiu, "DCAR: Distributed
Coding-Aware Routing in Wireless Networks," Distributed
Computing Systems, 2008. ICDCS '08. The 28th
International Conference on , vol., no., pp.462-469, 17-20
June 2008
Campista, M.E.M.; Esposito, P.M.; Moraes, I.M.; Costa,
L.H.M.; Duarte, O.C.M.; Passos, D.G.; de Albuquerque,
C.V.N.; Saade, D.C.M.; Rubinstein, M.G., "Routing
Metrics and Protocols for Wireless Mesh Networks,"
Network, IEEE , vol.22, no.1, pp.6-12, Jan.-Feb. 2008
Krishna Ramachandran, Irfan Sheriff, Elizabeth Belding
and Kevin Almeroth "A Multi-Radio 802.11 Mesh
Network Architecture" .Mobile Networks and Applications
Journal, Special Issue on Wireless Mesh Networks 2008
Mhatre, V., Lundgren, H., Baccelli, F., and Diot, C. 2007.
Joint MAC-aware routing and load balancing in mesh
networks. In Proceedings of the 2007 ACM CoNEXT
Conference (New York, New York, December 10 - 13,
2007). CoNEXT '07.
Gkantsidis, C., Hu, W., Key, P., Radunovic, B., Rodriguez,
P., and Gheorghiu, S. 2007. Multipath code casting for
wireless mesh networks. In Proceedings of the 2007 ACM
CoNEXT Conference (New York, New York, December 10
- 13, 2007). CoNEXT '07
D. Koutsonikolas, Y. Hu, and K. Papagiannaki
How
to
Evaluate
Exotic
Routing
Protocols
In ACM Hotnets, Calgary, Canada, October 2008.
De Couto, D. S., Aguayo, D., Bicket, J., and Morris, R.
2003. A high-throughput path metric for multi-hop
wireless routing. In Proceedings of the 9th Annual
international Conference on Mobile Computing and
Networking (San Diego, CA, USA, September 14 - 19,
2003). MobiCom '03. ACM, New York, NY, 134-146.
Juncheng Jia; Jin Zhang; Qian Zhang, "Relay-Assisted
Routing in Cognitive Radio Networks," Communications,
2009. ICC '09. IEEE International Conference on , vol.,
no., pp.1-5, 14-18 June 2009
I. Pefkianakis, S.H.Y. Wong, S. Lu, SAMER: Spectrum
Aware Mesh Routing in Cognitive Radio
Networks, DySPAN, 2008.
Jos Nez-Martnez was born in Barcelona,

Spain, in 1979. He received the degree in
Computer Science Engineering in 2004 from the
Technical University of Catalonia (UPC). He
worked as lab engineer in the Advanced
Broadband Communications Center (CCABA)
of Technical University of Catalonia (UPC) from 2003 to 2007.
Since 2007, he is working in Center of Technological
Telecommunications of Catalonia (CTTC) as a research
231
engineer. His research interests include the design and

implementation of routing protocols for wireless mesh
networks, and wireless network performance measurements.
Josep Mangues-Bafalluy received the degree
in Telecommunications Engineering in 1996,
and the PhD degree in Telecommunications in
2003, both from the Technical University of
Catalonia (UPC). Until 2003, he worked as a
researcher in the Broadband Communications
Research Group and the Advanced Broadband
Communications Center (CCABA) of UPC, and as an assistant
professor at UPC. Currently, he is a Senior Research Associate
of the Center of Technological Telecommunications of
Catalonia (CTTC) where he coordinates the activities of the IP
Technologies Area of the Centre. He has participated in several
Spanish (SABA1/2, SAM, ARTICO), EUREKA (Planets) and
EC-funded (LONG), WIP, N++) projects. His research interest
areas include mobility, self-organized networks and wireless
network performance measurements.

A Survey On Routing Protocols That Really Exploit Wireless Mesh Network Features

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Survey On Routing Protocols That Really Exploit Wireless Mesh Network Features

Uploaded by

Copyright:

Available Formats

JOURNAL OF COMMUNICATIONS, VOL. 5, NO.

A Survey on Routing Protocols that really Exploit

Centre Tecnolgic de Telecomunicacions de Catalunya (CTTC)

exploits the specific features of a Wireless Mesh

2010 ACADEMY PUBLISHER

in terms of connectivity: either the wireless link works

throughput. Therefore, the main goal of a routing

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

Furthermore, we are assuming all the WMRs in the

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

Input Multiple Output (MIMO) systems have been

Figure 1. WMN Architecture.

Multiple channels. An intelligent channel assignment

2010 ACADEMY PUBLISHER

routing protocol. For instance, if WMRs are equipped

words, there are some major choices to be made when

2010 ACADEMY PUBLISHER

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

and channel assignment becomes essential in multi-radio

Geographic routing. Scalability may be one of the

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

consequence of some properties present in a WMN. The

Previous classifications of routing protocols for

transmission range. However, for the same reason,

2010 ACADEMY PUBLISHER

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

As the routing protocol dictates the traffic load

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

However, the widespread deployment of directional

backbone, the number of potential receivers of a

Extremely Opportunistic Routing (ExOR) [16]

2010 ACADEMY PUBLISHER

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

the constraints imposed by transmission conflicts. They

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

Before forwarding data packets, the forwarders store

2010 ACADEMY PUBLISHER

opportunities in order to maximize the inter-flow network

in wireless link asymmetry. In practice, this means that

2010 ACADEMY PUBLISHER

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

during a given time period. The BER primary metric

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

Link Quality Estimates. The final goal of the

omnidirectional antennas becomes straightforward with

A list of the identified open research issues related to

measurements. On the other hand, RSSI cannot capture

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

sense, as WMNs offer a stable backbone, no additional

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

2010 ACADEMY PUBLISHER

Moreover, there is a route discovery component that

In both above approaches, the control message

fact, there are some strategies focused on reducing the

2010 ACADEMY PUBLISHER

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

OPEN RESEARCH ISSUES

A list of the identified open research issues related to

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010

leads to path inefficiencies. In general, this is not a

2010 ACADEMY PUBLISHER

On the other hand, the broadcast approach, changes the

JOURNAL OF COMMUNICATIONS, VOL. 5, NO. 3, MARCH 2010